[{"data":1,"prerenderedAt":815},["ShallowReactive",2],{"/en-us/blog/what-is-a-large-language-model-llm":3,"navigation-en-us":40,"banner-en-us":451,"footer-en-us":461,"blog-post-authors-en-us-Itzik Gan Baruch":701,"blog-related-posts-en-us-what-is-a-large-language-model-llm":715,"blog-promotions-en-us":753,"next-steps-en-us":805},{"id":4,"title":5,"authorSlugs":6,"authors":8,"body":10,"category":11,"categorySlug":11,"config":12,"content":16,"date":20,"description":17,"extension":24,"externalUrl":25,"featured":14,"heroImage":19,"isFeatured":14,"meta":26,"navigation":27,"path":28,"publishedDate":20,"rawbody":29,"seo":30,"slug":13,"stem":34,"tagSlugs":35,"tags":38,"template":15,"updatedDate":25,"__hash__":39},"blogPosts/en-us/blog/what-is-a-large-language-model-llm.yml","What is a large language model (LLM)?",[7],"itzik-gan-baruch",[9],"Itzik Gan Baruch","Large language models (LLMs) are revolutionizing DevOps and DevSecOps approaches by simplifying complex tasks, such as code creation, log analysis, and vulnerability detection.\n\nIn this article, you will learn how LLMs work, their practical applications, and the main challenges to overcome in order to fully harness their potential.\n\n## What is an LLM?\n\nLLMs are artificial intelligence (AI) systems that can process and generate text autonomously. They are trained by analyzing vast amounts of data from a variety of sources, enabling them to master the linguistic structures, contextual relationships, and nuances of language.\n\nLLMs are a major breakthrough in the field of AI. Their ability to process, generate, and interpret text relies on sophisticated machine learning and natural language processing (NLP) techniques. These systems do not just process individual words; they analyze complex sequences to capture the overall meaning, subtle contexts, and linguistic nuances.\n\n## How do LLMs work?\n\nTo better understand how they work, let's explore some of the key features of large language models.\n\n### Supervised and unsupervised learning\n\nLLMs are trained using two complementary approaches: supervised learning and unsupervised learning. These two approaches to machine learning maximize their ability to analyze and generate text.\n\n* **Supervised learning** relies on labeled data, where each input is associated with an expected output. The model learns to associate these inputs with the correct outputs by adjusting its internal parameters to reduce prediction errors. Through this approach, the model acquires precise knowledge about specific tasks, such as text classification or named entity recognition.\n\n* **Unsupervised learning (or machine learning)**, on the other hand, does not require labeled data. The model explores large volumes of text to discover hidden structures and identify semantic relationships. The model is therefore able to learn recurring patterns, implicit grammatical rules in the text, and contextualization of sentences and concepts. This method allows LLMs to be trained on large corpora of data, greatly accelerating their progress without direct human action.\n\nBy combining these two approaches, large language models gain the advantages of both precise, human-guided learning and unlimited autonomous exploration. This complementarity allows them to develop rapidly, while continuously improving their ability to understand and generate text coherently and contextually.\n\n### Learning based on a large volume of data\n\nLLMs are trained on billions of sentences from a variety of sources, such as news articles, online forums, technical documentation, scientific studies, and more. This variety of sources allows them to acquire a broad and nuanced understanding of natural language, ranging from everyday expressions to specialized terminology.\n\nThe richness of the data used is a key factor in LLMs' performance. Each source brings different writing styles, cultural contexts, and levels of technicality. \n\nFor example:\n\n* **News articles** to master informative and factual language \n* **Online forums** to understand specialized communities' informal conversations and technical language  \n* **Technical documentation and scientific studies** to assimilate complex concepts and specific terminology, particularly in areas such as DevOps and DevSecOps\n\nThis diversity of content allows LLMs to recognize complex linguistic structures, interpret sentences in different contexts, and adapt to highly technical domains. In DevSecOps, this means understanding commands, configurations, security protocols, and even concepts related to the development and maintenance of computer systems.\n\nWith this large-scale training, LLMs can accurately answer complex questions, write technical documentation, or identify vulnerabilities in computer systems.\n\n### Neural network architecture and \"deep learning\"\n\nLLMs are based on advanced neural network architectures. These networks are specially designed to process large sequences of text while maintaining an accurate understanding of the context. This deep learning-based training is a major asset in the field of NLP.\n\nThe best-known of these structures is the architecture of sequence-to-sequence models (transformers). This architecture has revolutionized NLP with its ability to simultaneously analyze all parts of a text, unlike sequential approaches that process words one by one.\n\nSequence-to-sequence models excel at processing long texts. For example, in a conversation or a detailed technical document, they are able to link distant information in the text to produce precise and well-reasoned answers. This context management is essential in a DevSecOps approach, where instructions can be complex and spread over multiple lines of code or configuration steps.\n\n### Predictive text generation\n\nWhen the user submits a text, query, or question, an LLM uses its predictive ability to generate the most likely sequence, based on the context provided.\n\nThe model analyzes each word, studies grammatical and semantic relationships, and then selects the most suitable terms to produce a coherent and informative text. This approach makes it possible to generate precise, detailed responses adapted to the expected tone.\n\nIn DevSecOps environments, this capability becomes particularly useful for:\n\n* **Coding assistance:** generation of code blocks or scripts adapted to specific configurations\n* **Technical problem solving:** proposing solutions based on descriptions of bugs or errors\n* **Drafting technical documentation:** automatic creation of guides, manuals, or instructions\n\nPredictive text generation thus makes it possible to automate many repetitive tasks and speed up technical teams' work.\n\n## Applications of large language models in a DevSecOps approach\n\nWith the rise of automation, LLMs have become indispensable allies for technical teams. Their ability to understand and generate text contextually enables them to effectively operate in complex environments such as [DevSecOps](https://about.gitlab.com/topics/devsecops/).\n\nWith their analytical power and ability to adapt to specific needs, these models offer tailored solutions to streamline processes and lighten technical teams' workload.\n\nDevelopment teams can leverage LLMs to automatically transform functional specifications into source code. \n\nWith this capability, they can perform the following actions:\n- generate complex automation scripts\n- create CI/CD pipelines tailored to specific business processes\n- produce customized security patches\n- generate code explanation and create documentation\n- refactor code by improving code structure and readability without changing functionality\n- generate tests\n\nBy relying on LLMs, teams are able to accelerate the development of their software while reducing the risk of human error.\n\n### Improved documentation and knowledge sharing\n\nThese powerful tools make it easy to create customized user manuals, API descriptions, and tutorials that are perfectly tailored to each user's level of expertise. By leveraging existing knowledge bases, LLMs create contextual answers to frequently asked questions. This enhances knowledge sharing within teams, speeds up onboarding of new members, and helps centralize best practices.\n\n### Incident management and troubleshooting\nDuring an incident, LLMs play a crucial role in analyzing logs and [trace files](https://docs.gitlab.com/development/tracing/) in real time. Thanks to their ability to cross-reference information from multiple sources, they identify anomalies and propose solutions based on similar past incidents. This approach significantly reduces diagnosis time. In addition, LLMs can automate the creation of detailed incident reports and recommend specific corrective actions.\n\n### Creating and improving CI/CD pipelines\n\nLLMs are revolutionizing the configuration of [CI/CD pipelines](https://about.gitlab.com/topics/ci-cd/cicd-pipeline/). They can not only help create pipelines, but also automate this process and suggest optimal configurations based on industry standards. By adapting workflows to your specific needs, they ensure perfect consistency between different development environments. Automated testing is enhanced by relevant suggestions, limiting the risk of failure. LLMs also continuously monitor the efficiency of pipelines and adjust processes to ensure smooth and uninterrupted rollout.\n\n### Security and compliance\n\nIn a DevSecOps environment, large language models become valuable allies for security and compliance. They parse the source code for potential vulnerabilities and generate detailed patch recommendations. LLMs can also monitor the application of security standards in real time, produce comprehensive compliance reports, and automate the application of security patches as soon as a vulnerability is identified. This automation enhances overall security and ensures consistent compliance with legal and industry requirements.\n\n## What are the benefits of large language models?\n\nLLMs are radically reshaping DevOps and DevSecOps approaches, bringing substantial improvements in productivity, security, and software quality. By integrating with existing workflows, LLMs are disrupting traditional approaches by automating complex tasks and providing innovative solutions.\n\n### Improved productivity and efficiency\n\nLLMs play a central role in improving technical teams' productivity and efficiency. By automating a wide range of repetitive tasks, they free development teams from routine operations, allowing them to focus on strategic activities with higher added value.\n\nIn addition, LLMs act as intelligent technical assistants capable of instantly providing relevant code snippets, tailored to the specific context of each project. In this way, they significantly reduce research time by offering ready-to-use solutions to assist teams in their work. This targeted assistance speeds up problem solving and reduces disruptions in workflows.\nAs a result, productivity increases and projects move forward more quickly. Technical teams can take on more tasks without compromising the quality of deliverables.\n\n### Improved code quality and security\n\nThe use of large language models in software development is a major lever for improving both code quality and application security. With their advanced analytical capabilities, LLMs can scan source code line by line and instantly detect syntax errors, logical inconsistencies, and potential vulnerabilities. Their ability to recognize defective code allows them to recommend appropriate fixes that comply with industry best practices.\n\nLLMs also play a key preventive role. They excel at identifying complex security flaws that are often difficult for humans to detect. By analyzing dependencies, they can flag obsolete or vulnerable libraries and recommend more secure, up-to-date versions. This approach contributes to maintaining a secure environment that complies with current security standards.\n\nBeyond fixing existing errors, LLMs offer improvements by suggesting optimized coding practices and project structures. They can generate code that meets the most advanced security standards from the earliest stages of development.\n\n### Accelerating development lifecycles\n\nLarge language models play a key role in accelerating software development lifecycles by automating key tasks that would otherwise tie up valuable human resources. Complex and repetitive tasks, such as writing functions, creating unit tests, or implementing standard components, are automated in a matter of moments.\n\nLLMs also speed up the validation phase with their ability to suggest complete and appropriate test cases. They ensure broader test coverage in less time, reducing the risk of errors and enabling early detection of anomalies. This preventive approach shortens the correction cycle and limits delays related to code quality issues.\n\nBy simplifying technical tasks and providing fast and tailored solutions, large language models enable businesses to respond to market demands in a more agile way. This acceleration of the development lifecycle results in more frequent updates, faster iterations, and a better ability to adapt products to users' changing needs.\n\nDevelopment lifecycles are becoming shorter, providing a critical strategic advantage in an increasingly demanding technology landscape.\n\n## What are the challenges of using LLMs?\n\nDespite their many benefits, large language models have certain limitations that require careful management. Their effectiveness depends heavily on the quality of the data used during their training and regular updates to their knowledge bases. In addition, issues related to algorithmic bias, data security, and privacy can arise, exposing companies to operational and legal risks. Rigorous human oversight remains essential in order to ensure the reliability of results, maintain regulatory compliance, and prevent critical errors.\n\n### Data privacy and security\n\nTraining LLMs relies on large volumes of data, often from diverse sources, raising questions about the protection of confidential information. Sensitive data shared with cloud platforms can therefore be exposed to potential breaches. This is of particular concern to companies operating in regulated sectors. \n\nIn Europe, where strict regulations like GDPR govern data management, many companies are reluctant to transfer their information to external services. Regulatory requirements, coupled with the fear of unauthorized exploitation of sensitive data, have led some companies to opt for self-hosted solutions to maintain complete control over their systems.\n\nProviders like GitLab have put in place robust security guarantees, such as intentional non-retention of personal data and end-to-end encryption. However, this may not be enough for the most demanding customers, who prefer complete control of their environments. Implementing hybrid or on-premises solutions then becomes a strategic necessity to meet the security requirements of certain companies.\n\nLearn more about GitLab Duo Self-Hosted by clicking on the image below to access our product tour.\n\n[![GitLab Duo Self-Hosted tour](https://res.cloudinary.com/about-gitlab-com/image/upload/v1749673815/Blog/Content%20Images/Screenshot_2025-05-29_at_8.29.30%C3%A2__AM.png)](https://gitlab.navattic.com/gitlab-duo-self-hosted)\n\n### Accuracy and reliability\n\nAlthough large language models are capable of producing impressive results, their performance is not infallible. They can produce incorrect, incomplete, or inconsistent answers. This inaccuracy becomes particularly problematic in the context of critical tasks such as generating security code or analyzing sensitive data.\n\nIn addition, LLMs operate on the basis of probabilistic models, which means that they do not truly \"understand\" the content they process, but produce predictions based on statistical probabilities. This can lead to technically incorrect or even dangerous recommendations when used without human validation.\n\nTo avoid these pitfalls, it is essential to maintain constant oversight and establish rigorous validation processes. The results provided by LLMs must always be reviewed by humans before being integrated into critical systems.\n\nA strategy of regular model updates, combined with proactive human oversight, can reduce errors and gradually improve the reliability of results.\n\n## How GitLab uses LLMs for GitLab Duo features\n\n[GitLab Duo](https://about.gitlab.com/gitlab-duo-agent-platform/) harnesses the power of large language models to transform DevSecOps processes by integrating AI-powered capabilities throughout the software development lifecycle. This approach aims to improve productivity, strengthen security, and automate complex tasks so that development teams can focus on high added-value tasks.\n\n### AI-assisted software development\n\nGitLab Duo provides continuous support throughout the software development lifecycle with real-time recommendations. Development teams can automatically generate unit tests, get detailed explanations of complex code segments, and benefit from suggestions to improve the quality of their code.\n\n### Proactive CI/CD failure analysis\n\nOne of the key features of GitLab Duo is its assistance in analyzing CI/CD job failures. With LLM and AI, teams are able to quickly identify sources of errors in their continuous integration and deployment pipelines. \n\n### Enhanced code security\n\nGitLab Duo incorporates AI-based security features. The system detects vulnerabilities in the source code and proposes detailed patches to reduce the risks. Teams receive clear explanations of the nature of the vulnerabilities identified and can apply automated patches via [merge requests](https://docs.gitlab.com/user/project/merge_requests/) generated directly by GitLab Duo. This feature helps secure development without slowing down development lifecycles.\n\nLearn more about GitLab Duo Vulnerability Explanation and Resolution by clicking on the image below to access our product tour.\n\n[![Vulnerability report interactive tour](https://res.cloudinary.com/about-gitlab-com/image/upload/v1749673816/Blog/Content%20Images/Screenshot_2025-05-29_at_8.32.15%C3%A2__AM.png)](https://gitlab.navattic.com/ve-vr-short)\n\n### Key features of GitLab Duo\n\n* [GitLab Duo Chat](https://about.gitlab.com/blog/10-best-practices-for-using-ai-powered-gitlab-duo-chat/): This conversational feature processes and generates text and code intuitively. It allows users to quickly search for relevant information in large volumes of text, including in tickets, [epics](https://docs.gitlab.com/user/group/epics/), source code, and [GitLab documentation](https://docs.gitlab.com/).\n\n* [GitLab Duo Self-Hosted](https://about.gitlab.com/blog/gitlab-duo-self-hosted-enterprise-ai-built-for-data-privacy/): GitLab Duo Self-Hosted allows companies with strict data privacy requirements to benefit from GitLab Duo's AI capabilities with flexibility in choosing deployment and LLMs from a list of supported options.\n\n* [GitLab Duo Code Suggestions](https://docs.gitlab.com/user/project/repository/code_suggestions/): Development teams benefit from automated code suggestions, allowing them to write secure code faster. Repetitive and routine coding tasks are automated, significantly speeding up software development lifecycles.\n\nGitLab Duo is not limited to these features. It offers a wide range of features designed to simplify and optimize software development. Whether it's automating testing, improving collaboration between teams, or strengthening project security, GitLab Duo is a complete solution for smart and efficient DevSecOps processes.\n\nLearn more about GitLab Duo Enterprise by clicking on the image below to access our product tour. \n\n[![GitLab Duo Enterprise interactive tour](https://res.cloudinary.com/about-gitlab-com/image/upload/v1749673816/Blog/Content%20Images/Screenshot_2025-05-29_at_8.33.40%C3%A2__AM.png)](https://gitlab.navattic.com/duo-enterprise)","ai-ml",{"slug":13,"featured":14,"template":15},"what-is-a-large-language-model-llm",false,"BlogPost",{"title":5,"description":17,"authors":18,"heroImage":19,"date":20,"body":10,"category":11,"tags":21},"Learn how large language models work, their applications, and their impact on the DevSecOps world.",[9],"https://res.cloudinary.com/about-gitlab-com/image/upload/v1749660057/Blog/Hero%20Images/LLM.jpg","2025-05-29",[22,23],"AI/ML","DevSecOps","yml",null,{},true,"/en-us/blog/what-is-a-large-language-model-llm","seo:\n  title: What is a large language model (LLM)?\n  description: >-\n    Learn how large language models work, their applications, and their impact\n    on the DevSecOps world.\n  ogTitle: What is a large language model (LLM)?\n  ogDescription: >-\n    Learn how large language models work, their applications, and their impact\n    on the DevSecOps world.\n  noIndex: false\n  ogImage: >-\n    https://res.cloudinary.com/about-gitlab-com/image/upload/v1749660057/Blog/Hero%20Images/LLM.jpg\n  ogUrl: https://about.gitlab.com/blog/what-is-a-large-language-model-llm\n  ogSiteName: https://about.gitlab.com\n  ogType: article\n  canonicalUrls: https://about.gitlab.com/blog/what-is-a-large-language-model-llm\ncontent:\n  title: What is a large language model (LLM)?\n  description: >-\n    Learn how large language models work, their applications, and their impact\n    on the DevSecOps world.\n  authors:\n    - Itzik Gan Baruch\n  heroImage: >-\n    https://res.cloudinary.com/about-gitlab-com/image/upload/v1749660057/Blog/Hero%20Images/LLM.jpg\n  date: '2025-05-29'\n  body: >-\n    Large language models (LLMs) are revolutionizing DevOps and DevSecOps\n    approaches by simplifying complex tasks, such as code creation, log\n    analysis, and vulnerability detection.\n\n\n    In this article, you will learn how LLMs work, their practical applications,\n    and the main challenges to overcome in order to fully harness their\n    potential.\n\n\n    ## What is an LLM?\n\n\n    LLMs are artificial intelligence (AI) systems that can process and generate\n    text autonomously. They are trained by analyzing vast amounts of data from a\n    variety of sources, enabling them to master the linguistic structures,\n    contextual relationships, and nuances of language.\n\n\n    LLMs are a major breakthrough in the field of AI. Their ability to process,\n    generate, and interpret text relies on sophisticated machine learning and\n    natural language processing (NLP) techniques. These systems do not just\n    process individual words; they analyze complex sequences to capture the\n    overall meaning, subtle contexts, and linguistic nuances.\n\n\n    ## How do LLMs work?\n\n\n    To better understand how they work, let's explore some of the key features\n    of large language models.\n\n\n    ### Supervised and unsupervised learning\n\n\n    LLMs are trained using two complementary approaches: supervised learning and\n    unsupervised learning. These two approaches to machine learning maximize\n    their ability to analyze and generate text.\n\n\n    * **Supervised learning** relies on labeled data, where each input is\n    associated with an expected output. The model learns to associate these\n    inputs with the correct outputs by adjusting its internal parameters to\n    reduce prediction errors. Through this approach, the model acquires precise\n    knowledge about specific tasks, such as text classification or named entity\n    recognition.\n\n\n    * **Unsupervised learning (or machine learning)**, on the other hand, does\n    not require labeled data. The model explores large volumes of text to\n    discover hidden structures and identify semantic relationships. The model is\n    therefore able to learn recurring patterns, implicit grammatical rules in\n    the text, and contextualization of sentences and concepts. This method\n    allows LLMs to be trained on large corpora of data, greatly accelerating\n    their progress without direct human action.\n\n\n    By combining these two approaches, large language models gain the advantages\n    of both precise, human-guided learning and unlimited autonomous exploration.\n    This complementarity allows them to develop rapidly, while continuously\n    improving their ability to understand and generate text coherently and\n    contextually.\n\n\n    ### Learning based on a large volume of data\n\n\n    LLMs are trained on billions of sentences from a variety of sources, such as\n    news articles, online forums, technical documentation, scientific studies,\n    and more. This variety of sources allows them to acquire a broad and nuanced\n    understanding of natural language, ranging from everyday expressions to\n    specialized terminology.\n\n\n    The richness of the data used is a key factor in LLMs' performance. Each\n    source brings different writing styles, cultural contexts, and levels of\n    technicality. \n\n\n    For example:\n\n\n    * **News articles** to master informative and factual language \n\n    * **Online forums** to understand specialized communities' informal\n    conversations and technical language  \n\n    * **Technical documentation and scientific studies** to assimilate complex\n    concepts and specific terminology, particularly in areas such as DevOps and\n    DevSecOps\n\n\n    This diversity of content allows LLMs to recognize complex linguistic\n    structures, interpret sentences in different contexts, and adapt to highly\n    technical domains. In DevSecOps, this means understanding commands,\n    configurations, security protocols, and even concepts related to the\n    development and maintenance of computer systems.\n\n\n    With this large-scale training, LLMs can accurately answer complex\n    questions, write technical documentation, or identify vulnerabilities in\n    computer systems.\n\n\n    ### Neural network architecture and \"deep learning\"\n\n\n    LLMs are based on advanced neural network architectures. These networks are\n    specially designed to process large sequences of text while maintaining an\n    accurate understanding of the context. This deep learning-based training is\n    a major asset in the field of NLP.\n\n\n    The best-known of these structures is the architecture of\n    sequence-to-sequence models (transformers). This architecture has\n    revolutionized NLP with its ability to simultaneously analyze all parts of a\n    text, unlike sequential approaches that process words one by one.\n\n\n    Sequence-to-sequence models excel at processing long texts. For example, in\n    a conversation or a detailed technical document, they are able to link\n    distant information in the text to produce precise and well-reasoned\n    answers. This context management is essential in a DevSecOps approach, where\n    instructions can be complex and spread over multiple lines of code or\n    configuration steps.\n\n\n    ### Predictive text generation\n\n\n    When the user submits a text, query, or question, an LLM uses its predictive\n    ability to generate the most likely sequence, based on the context provided.\n\n\n    The model analyzes each word, studies grammatical and semantic\n    relationships, and then selects the most suitable terms to produce a\n    coherent and informative text. This approach makes it possible to generate\n    precise, detailed responses adapted to the expected tone.\n\n\n    In DevSecOps environments, this capability becomes particularly useful for:\n\n\n    * **Coding assistance:** generation of code blocks or scripts adapted to\n    specific configurations\n\n    * **Technical problem solving:** proposing solutions based on descriptions\n    of bugs or errors\n\n    * **Drafting technical documentation:** automatic creation of guides,\n    manuals, or instructions\n\n\n    Predictive text generation thus makes it possible to automate many\n    repetitive tasks and speed up technical teams' work.\n\n\n    ## Applications of large language models in a DevSecOps approach\n\n\n    With the rise of automation, LLMs have become indispensable allies for\n    technical teams. Their ability to understand and generate text contextually\n    enables them to effectively operate in complex environments such as\n    [DevSecOps](https://about.gitlab.com/topics/devsecops/).\n\n\n    With their analytical power and ability to adapt to specific needs, these\n    models offer tailored solutions to streamline processes and lighten\n    technical teams' workload.\n\n\n    Development teams can leverage LLMs to automatically transform functional\n    specifications into source code. \n\n\n    With this capability, they can perform the following actions:\n\n    - generate complex automation scripts\n\n    - create CI/CD pipelines tailored to specific business processes\n\n    - produce customized security patches\n\n    - generate code explanation and create documentation\n\n    - refactor code by improving code structure and readability without changing\n    functionality\n\n    - generate tests\n\n\n    By relying on LLMs, teams are able to accelerate the development of their\n    software while reducing the risk of human error.\n\n\n    ### Improved documentation and knowledge sharing\n\n\n    These powerful tools make it easy to create customized user manuals, API\n    descriptions, and tutorials that are perfectly tailored to each user's level\n    of expertise. By leveraging existing knowledge bases, LLMs create contextual\n    answers to frequently asked questions. This enhances knowledge sharing\n    within teams, speeds up onboarding of new members, and helps centralize best\n    practices.\n\n\n    ### Incident management and troubleshooting\n\n    During an incident, LLMs play a crucial role in analyzing logs and [trace\n    files](https://docs.gitlab.com/development/tracing/) in real time.\n    Thanks to their ability to cross-reference information from multiple\n    sources, they identify anomalies and propose solutions based on similar past\n    incidents. This approach significantly reduces diagnosis time. In addition,\n    LLMs can automate the creation of detailed incident reports and recommend\n    specific corrective actions.\n\n\n    ### Creating and improving CI/CD pipelines\n\n\n    LLMs are revolutionizing the configuration of [CI/CD\n    pipelines](https://about.gitlab.com/topics/ci-cd/cicd-pipeline/). They can\n    not only help create pipelines, but also automate this process and suggest\n    optimal configurations based on industry standards. By adapting workflows to\n    your specific needs, they ensure perfect consistency between different\n    development environments. Automated testing is enhanced by relevant\n    suggestions, limiting the risk of failure. LLMs also continuously monitor\n    the efficiency of pipelines and adjust processes to ensure smooth and\n    uninterrupted rollout.\n\n\n    ### Security and compliance\n\n\n    In a DevSecOps environment, large language models become valuable allies for\n    security and compliance. They parse the source code for potential\n    vulnerabilities and generate detailed patch recommendations. LLMs can also\n    monitor the application of security standards in real time, produce\n    comprehensive compliance reports, and automate the application of security\n    patches as soon as a vulnerability is identified. This automation enhances\n    overall security and ensures consistent compliance with legal and industry\n    requirements.\n\n\n    ## What are the benefits of large language models?\n\n\n    LLMs are radically reshaping DevOps and DevSecOps approaches, bringing\n    substantial improvements in productivity, security, and software quality. By\n    integrating with existing workflows, LLMs are disrupting traditional\n    approaches by automating complex tasks and providing innovative solutions.\n\n\n    ### Improved productivity and efficiency\n\n\n    LLMs play a central role in improving technical teams' productivity and\n    efficiency. By automating a wide range of repetitive tasks, they free\n    development teams from routine operations, allowing them to focus on\n    strategic activities with higher added value.\n\n\n    In addition, LLMs act as intelligent technical assistants capable of\n    instantly providing relevant code snippets, tailored to the specific context\n    of each project. In this way, they significantly reduce research time by\n    offering ready-to-use solutions to assist teams in their work. This targeted\n    assistance speeds up problem solving and reduces disruptions in workflows.\n\n    As a result, productivity increases and projects move forward more quickly.\n    Technical teams can take on more tasks without compromising the quality of\n    deliverables.\n\n\n    ### Improved code quality and security\n\n\n    The use of large language models in software development is a major lever\n    for improving both code quality and application security. With their\n    advanced analytical capabilities, LLMs can scan source code line by line and\n    instantly detect syntax errors, logical inconsistencies, and potential\n    vulnerabilities. Their ability to recognize defective code allows them to\n    recommend appropriate fixes that comply with industry best practices.\n\n\n    LLMs also play a key preventive role. They excel at identifying complex\n    security flaws that are often difficult for humans to detect. By analyzing\n    dependencies, they can flag obsolete or vulnerable libraries and recommend\n    more secure, up-to-date versions. This approach contributes to maintaining a\n    secure environment that complies with current security standards.\n\n\n    Beyond fixing existing errors, LLMs offer improvements by suggesting\n    optimized coding practices and project structures. They can generate code\n    that meets the most advanced security standards from the earliest stages of\n    development.\n\n\n    ### Accelerating development lifecycles\n\n\n    Large language models play a key role in accelerating software development\n    lifecycles by automating key tasks that would otherwise tie up valuable\n    human resources. Complex and repetitive tasks, such as writing functions,\n    creating unit tests, or implementing standard components, are automated in a\n    matter of moments.\n\n\n    LLMs also speed up the validation phase with their ability to suggest\n    complete and appropriate test cases. They ensure broader test coverage in\n    less time, reducing the risk of errors and enabling early detection of\n    anomalies. This preventive approach shortens the correction cycle and limits\n    delays related to code quality issues.\n\n\n    By simplifying technical tasks and providing fast and tailored solutions,\n    large language models enable businesses to respond to market demands in a\n    more agile way. This acceleration of the development lifecycle results in\n    more frequent updates, faster iterations, and a better ability to adapt\n    products to users' changing needs.\n\n\n    Development lifecycles are becoming shorter, providing a critical strategic\n    advantage in an increasingly demanding technology landscape.\n\n\n    ## What are the challenges of using LLMs?\n\n\n    Despite their many benefits, large language models have certain limitations\n    that require careful management. Their effectiveness depends heavily on the\n    quality of the data used during their training and regular updates to their\n    knowledge bases. In addition, issues related to algorithmic bias, data\n    security, and privacy can arise, exposing companies to operational and legal\n    risks. Rigorous human oversight remains essential in order to ensure the\n    reliability of results, maintain regulatory compliance, and prevent critical\n    errors.\n\n\n    ### Data privacy and security\n\n\n    Training LLMs relies on large volumes of data, often from diverse sources,\n    raising questions about the protection of confidential information.\n    Sensitive data shared with cloud platforms can therefore be exposed to\n    potential breaches. This is of particular concern to companies operating in\n    regulated sectors. \n\n\n    In Europe, where strict regulations like GDPR govern data management, many\n    companies are reluctant to transfer their information to external services.\n    Regulatory requirements, coupled with the fear of unauthorized exploitation\n    of sensitive data, have led some companies to opt for self-hosted solutions\n    to maintain complete control over their systems.\n\n\n    Providers like GitLab have put in place robust security guarantees, such as\n    intentional non-retention of personal data and end-to-end encryption.\n    However, this may not be enough for the most demanding customers, who prefer\n    complete control of their environments. Implementing hybrid or on-premises\n    solutions then becomes a strategic necessity to meet the security\n    requirements of certain companies.\n\n\n    Learn more about GitLab Duo Self-Hosted by clicking on the image below to\n    access our product tour.\n\n\n    [![GitLab Duo Self-Hosted\n    tour](https://res.cloudinary.com/about-gitlab-com/image/upload/v1749673815/Blog/Content%20Images/Screenshot_2025-05-29_at_8.29.30%C3%A2__AM.png)](https://gitlab.navattic.com/gitlab-duo-self-hosted)\n\n\n    ### Accuracy and reliability\n\n\n    Although large language models are capable of producing impressive results,\n    their performance is not infallible. They can produce incorrect, incomplete,\n    or inconsistent answers. This inaccuracy becomes particularly problematic in\n    the context of critical tasks such as generating security code or analyzing\n    sensitive data.\n\n\n    In addition, LLMs operate on the basis of probabilistic models, which means\n    that they do not truly \"understand\" the content they process, but produce\n    predictions based on statistical probabilities. This can lead to technically\n    incorrect or even dangerous recommendations when used without human\n    validation.\n\n\n    To avoid these pitfalls, it is essential to maintain constant oversight and\n    establish rigorous validation processes. The results provided by LLMs must\n    always be reviewed by humans before being integrated into critical systems.\n\n\n    A strategy of regular model updates, combined with proactive human\n    oversight, can reduce errors and gradually improve the reliability of\n    results.\n\n\n    ## How GitLab uses LLMs for GitLab Duo features\n\n\n    [GitLab Duo](https://about.gitlab.com/gitlab-duo-agent-platform/) harnesses the power of\n    large language models to transform DevSecOps processes by integrating\n    AI-powered capabilities throughout the software development lifecycle. This\n    approach aims to improve productivity, strengthen security, and automate\n    complex tasks so that development teams can focus on high added-value tasks.\n\n\n    ### AI-assisted software development\n\n\n    GitLab Duo provides continuous support throughout the software development\n    lifecycle with real-time recommendations. Development teams can\n    automatically generate unit tests, get detailed explanations of complex code\n    segments, and benefit from suggestions to improve the quality of their code.\n\n\n    ### Proactive CI/CD failure analysis\n\n\n    One of the key features of GitLab Duo is its assistance in analyzing CI/CD\n    job failures. With LLM and AI, teams are able to quickly identify sources of\n    errors in their continuous integration and deployment pipelines. \n\n\n    ### Enhanced code security\n\n\n    GitLab Duo incorporates AI-based security features. The system detects\n    vulnerabilities in the source code and proposes detailed patches to reduce\n    the risks. Teams receive clear explanations of the nature of the\n    vulnerabilities identified and can apply automated patches via [merge\n    requests](https://docs.gitlab.com/user/project/merge_requests/) generated\n    directly by GitLab Duo. This feature helps secure development without\n    slowing down development lifecycles.\n\n\n    Learn more about GitLab Duo Vulnerability Explanation and Resolution by\n    clicking on the image below to access our product tour.\n\n\n    [![Vulnerability report interactive\n    tour](https://res.cloudinary.com/about-gitlab-com/image/upload/v1749673816/Blog/Content%20Images/Screenshot_2025-05-29_at_8.32.15%C3%A2__AM.png)](https://gitlab.navattic.com/ve-vr-short)\n\n\n    ### Key features of GitLab Duo\n\n\n    * [GitLab Duo\n    Chat](https://about.gitlab.com/blog/10-best-practices-for-using-ai-powered-gitlab-duo-chat/):\n    This conversational feature processes and generates text and code\n    intuitively. It allows users to quickly search for relevant information in\n    large volumes of text, including in tickets,\n    [epics](https://docs.gitlab.com/user/group/epics/), source code, and\n    [GitLab documentation](https://docs.gitlab.com/).\n\n\n    * [GitLab Duo\n    Self-Hosted](https://about.gitlab.com/blog/gitlab-duo-self-hosted-enterprise-ai-built-for-data-privacy/):\n    GitLab Duo Self-Hosted allows companies with strict data privacy\n    requirements to benefit from GitLab Duo's AI capabilities with flexibility\n    in choosing deployment and LLMs from a list of supported options.\n\n\n    * [GitLab Duo Code Suggestions](https://docs.gitlab.com/user/project/repository/code_suggestions/):\n    Development teams benefit from automated code suggestions, allowing them to\n    write secure code faster. Repetitive and routine coding tasks are automated,\n    significantly speeding up software development lifecycles.\n\n\n    GitLab Duo is not limited to these features. It offers a wide range of\n    features designed to simplify and optimize software development. Whether\n    it's automating testing, improving collaboration between teams, or\n    strengthening project security, GitLab Duo is a complete solution for smart\n    and efficient DevSecOps processes.\n\n\n    Learn more about GitLab Duo Enterprise by clicking on the image below to\n    access our product tour. \n\n\n    [![GitLab Duo Enterprise interactive\n    tour](https://res.cloudinary.com/about-gitlab-com/image/upload/v1749673816/Blog/Content%20Images/Screenshot_2025-05-29_at_8.33.40%C3%A2__AM.png)](https://gitlab.navattic.com/duo-enterprise)\n  category: ai-ml\n  tags:\n    - AI/ML\n    - DevSecOps\nconfig:\n  slug: what-is-a-large-language-model-llm\n  featured: false\n  template: BlogPost\n",{"title":5,"description":17,"ogTitle":5,"ogDescription":17,"noIndex":14,"ogImage":19,"ogUrl":31,"ogSiteName":32,"ogType":33,"canonicalUrls":31},"https://about.gitlab.com/blog/what-is-a-large-language-model-llm","https://about.gitlab.com","article","en-us/blog/what-is-a-large-language-model-llm",[36,37],"aiml","devsecops",[22,23],"TMZmhwNTmVL15PKZ__0xnI8jaMdhECd6jYLMCefBX2g",{"data":41},{"logo":42,"freeTrial":47,"sales":52,"login":57,"items":62,"search":371,"minimal":402,"duo":421,"switchNav":430,"pricingDeployment":441},{"config":43},{"href":44,"dataGaName":45,"dataGaLocation":46},"/","gitlab logo","header",{"text":48,"config":49},"Get free trial",{"href":50,"dataGaName":51,"dataGaLocation":46},"https://gitlab.com/-/trial_registrations/new?glm_source=about.gitlab.com&glm_content=default-saas-trial/","free trial",{"text":53,"config":54},"Talk to sales",{"href":55,"dataGaName":56,"dataGaLocation":46},"/sales/","sales",{"text":58,"config":59},"Sign in",{"href":60,"dataGaName":61,"dataGaLocation":46},"https://gitlab.com/users/sign_in/","sign in",[63,90,185,190,292,352],{"text":64,"config":65,"cards":67},"Platform",{"dataNavLevelOne":66},"platform",[68,74,82],{"title":64,"description":69,"link":70},"The intelligent orchestration platform for DevSecOps",{"text":71,"config":72},"Explore our Platform",{"href":73,"dataGaName":66,"dataGaLocation":46},"/platform/",{"title":75,"description":76,"link":77},"GitLab Duo Agent Platform","Agentic AI for the entire software lifecycle",{"text":78,"config":79},"Meet GitLab Duo",{"href":80,"dataGaName":81,"dataGaLocation":46},"/gitlab-duo-agent-platform/","gitlab duo agent platform",{"title":83,"description":84,"link":85},"Why GitLab","See the top reasons enterprises choose GitLab",{"text":86,"config":87},"Learn more",{"href":88,"dataGaName":89,"dataGaLocation":46},"/why-gitlab/","why gitlab",{"text":91,"left":27,"config":92,"link":94,"lists":98,"footer":167},"Product",{"dataNavLevelOne":93},"solutions",{"text":95,"config":96},"View all Solutions",{"href":97,"dataGaName":93,"dataGaLocation":46},"/solutions/",[99,123,146],{"title":100,"description":101,"link":102,"items":107},"Automation","CI/CD and automation to accelerate deployment",{"config":103},{"icon":104,"href":105,"dataGaName":106,"dataGaLocation":46},"AutomatedCodeAlt","/solutions/delivery-automation/","automated software delivery",[108,112,115,119],{"text":109,"config":110},"CI/CD",{"href":111,"dataGaLocation":46,"dataGaName":109},"/solutions/continuous-integration/",{"text":75,"config":113},{"href":80,"dataGaLocation":46,"dataGaName":114},"gitlab duo agent platform - product menu",{"text":116,"config":117},"Source Code Management",{"href":118,"dataGaLocation":46,"dataGaName":116},"/solutions/source-code-management/",{"text":120,"config":121},"Automated Software Delivery",{"href":105,"dataGaLocation":46,"dataGaName":122},"Automated software delivery",{"title":124,"description":125,"link":126,"items":131},"Security","Deliver code faster without compromising security",{"config":127},{"href":128,"dataGaName":129,"dataGaLocation":46,"icon":130},"/solutions/application-security-testing/","security and compliance","ShieldCheckLight",[132,136,141],{"text":133,"config":134},"Application Security Testing",{"href":128,"dataGaName":135,"dataGaLocation":46},"Application security testing",{"text":137,"config":138},"Software Supply Chain Security",{"href":139,"dataGaLocation":46,"dataGaName":140},"/solutions/supply-chain/","Software supply chain security",{"text":142,"config":143},"Software Compliance",{"href":144,"dataGaName":145,"dataGaLocation":46},"/solutions/software-compliance/","software compliance",{"title":147,"link":148,"items":153},"Measurement",{"config":149},{"icon":150,"href":151,"dataGaName":152,"dataGaLocation":46},"DigitalTransformation","/solutions/visibility-measurement/","visibility and measurement",[154,158,162],{"text":155,"config":156},"Visibility & Measurement",{"href":151,"dataGaLocation":46,"dataGaName":157},"Visibility and Measurement",{"text":159,"config":160},"Value Stream Management",{"href":161,"dataGaLocation":46,"dataGaName":159},"/solutions/value-stream-management/",{"text":163,"config":164},"Analytics & Insights",{"href":165,"dataGaLocation":46,"dataGaName":166},"/solutions/analytics-and-insights/","Analytics and insights",{"title":168,"items":169},"GitLab for",[170,175,180],{"text":171,"config":172},"Enterprise",{"href":173,"dataGaLocation":46,"dataGaName":174},"/enterprise/","enterprise",{"text":176,"config":177},"Small Business",{"href":178,"dataGaLocation":46,"dataGaName":179},"/small-business/","small business",{"text":181,"config":182},"Public Sector",{"href":183,"dataGaLocation":46,"dataGaName":184},"/solutions/public-sector/","public sector",{"text":186,"config":187},"Pricing",{"href":188,"dataGaName":189,"dataGaLocation":46,"dataNavLevelOne":189},"/pricing/","pricing",{"text":191,"config":192,"link":194,"lists":198,"feature":283},"Resources",{"dataNavLevelOne":193},"resources",{"text":195,"config":196},"View all resources",{"href":197,"dataGaName":193,"dataGaLocation":46},"/resources/",[199,232,255],{"title":200,"items":201},"Getting started",[202,207,212,217,222,227],{"text":203,"config":204},"Install",{"href":205,"dataGaName":206,"dataGaLocation":46},"/install/","install",{"text":208,"config":209},"Quick start guides",{"href":210,"dataGaName":211,"dataGaLocation":46},"/get-started/","quick setup checklists",{"text":213,"config":214},"Learn",{"href":215,"dataGaLocation":46,"dataGaName":216},"https://university.gitlab.com/","learn",{"text":218,"config":219},"Product documentation",{"href":220,"dataGaName":221,"dataGaLocation":46},"https://docs.gitlab.com/","product documentation",{"text":223,"config":224},"Best practice videos",{"href":225,"dataGaName":226,"dataGaLocation":46},"/getting-started-videos/","best practice videos",{"text":228,"config":229},"Integrations",{"href":230,"dataGaName":231,"dataGaLocation":46},"/integrations/","integrations",{"title":233,"items":234},"Discover",[235,240,245,250],{"text":236,"config":237},"Customer success stories",{"href":238,"dataGaName":239,"dataGaLocation":46},"/customers/","customer success stories",{"text":241,"config":242},"Blog",{"href":243,"dataGaName":244,"dataGaLocation":46},"/blog/","blog",{"text":246,"config":247},"The Source",{"href":248,"dataGaName":249,"dataGaLocation":46},"/the-source/","the source",{"text":251,"config":252},"Remote",{"href":253,"dataGaName":254,"dataGaLocation":46},"https://handbook.gitlab.com/handbook/company/culture/all-remote/","remote",{"title":256,"items":257},"Connect",[258,263,268,273,278],{"text":259,"config":260},"GitLab Services",{"href":261,"dataGaName":262,"dataGaLocation":46},"/services/","services",{"text":264,"config":265},"Community",{"href":266,"dataGaName":267,"dataGaLocation":46},"/community/","community",{"text":269,"config":270},"Forum",{"href":271,"dataGaName":272,"dataGaLocation":46},"https://forum.gitlab.com/","forum",{"text":274,"config":275},"Events",{"href":276,"dataGaName":277,"dataGaLocation":46},"/events/","events",{"text":279,"config":280},"Partners",{"href":281,"dataGaName":282,"dataGaLocation":46},"/partners/","partners",{"textColor":284,"title":285,"text":286,"link":287},"#000","What’s new in GitLab","Stay updated with our latest features and improvements.",{"text":288,"config":289},"Read the latest",{"href":290,"dataGaName":291,"dataGaLocation":46},"/releases/whats-new/","whats new",{"text":293,"config":294,"lists":296},"Company",{"dataNavLevelOne":295},"company",[297],{"items":298},[299,304,310,312,317,322,327,332,337,342,347],{"text":300,"config":301},"About",{"href":302,"dataGaName":303,"dataGaLocation":46},"/company/","about",{"text":305,"config":306,"footerGa":309},"Jobs",{"href":307,"dataGaName":308,"dataGaLocation":46},"/jobs/","jobs",{"dataGaName":308},{"text":274,"config":311},{"href":276,"dataGaName":277,"dataGaLocation":46},{"text":313,"config":314},"Leadership",{"href":315,"dataGaName":316,"dataGaLocation":46},"/company/team/e-group/","leadership",{"text":318,"config":319},"Team",{"href":320,"dataGaName":321,"dataGaLocation":46},"/company/team/","team",{"text":323,"config":324},"Handbook",{"href":325,"dataGaName":326,"dataGaLocation":46},"https://handbook.gitlab.com/","handbook",{"text":328,"config":329},"Investor relations",{"href":330,"dataGaName":331,"dataGaLocation":46},"https://ir.gitlab.com/","investor relations",{"text":333,"config":334},"Trust Center",{"href":335,"dataGaName":336,"dataGaLocation":46},"/security/","trust center",{"text":338,"config":339},"AI Transparency Center",{"href":340,"dataGaName":341,"dataGaLocation":46},"/ai-transparency-center/","ai transparency center",{"text":343,"config":344},"Newsletter",{"href":345,"dataGaName":346,"dataGaLocation":46},"/company/contact/#contact-forms","newsletter",{"text":348,"config":349},"Press",{"href":350,"dataGaName":351,"dataGaLocation":46},"/press/","press",{"text":353,"config":354,"lists":355},"Contact us",{"dataNavLevelOne":295},[356],{"items":357},[358,361,366],{"text":53,"config":359},{"href":55,"dataGaName":360,"dataGaLocation":46},"talk to sales",{"text":362,"config":363},"Support portal",{"href":364,"dataGaName":365,"dataGaLocation":46},"https://support.gitlab.com","support portal",{"text":367,"config":368},"Customer portal",{"href":369,"dataGaName":370,"dataGaLocation":46},"https://customers.gitlab.com/customers/sign_in/","customer portal",{"close":372,"login":373,"suggestions":380},"Close",{"text":374,"link":375},"To search repositories and projects, login to",{"text":376,"config":377},"gitlab.com",{"href":60,"dataGaName":378,"dataGaLocation":379},"search login","search",{"text":381,"default":382},"Suggestions",[383,385,389,391,395,399],{"text":75,"config":384},{"href":80,"dataGaName":75,"dataGaLocation":379},{"text":386,"config":387},"Code Suggestions (AI)",{"href":388,"dataGaName":386,"dataGaLocation":379},"/solutions/code-suggestions/",{"text":109,"config":390},{"href":111,"dataGaName":109,"dataGaLocation":379},{"text":392,"config":393},"GitLab on AWS",{"href":394,"dataGaName":392,"dataGaLocation":379},"/partners/technology-partners/aws/",{"text":396,"config":397},"GitLab on Google Cloud",{"href":398,"dataGaName":396,"dataGaLocation":379},"/partners/technology-partners/google-cloud-platform/",{"text":400,"config":401},"Why GitLab?",{"href":88,"dataGaName":400,"dataGaLocation":379},{"freeTrial":403,"mobileIcon":408,"desktopIcon":413,"secondaryButton":416},{"text":404,"config":405},"Start free trial",{"href":406,"dataGaName":51,"dataGaLocation":407},"https://gitlab.com/-/trials/new/","nav",{"altText":409,"config":410},"Gitlab Icon",{"src":411,"dataGaName":412,"dataGaLocation":407},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1758203874/jypbw1jx72aexsoohd7x.svg","gitlab icon",{"altText":409,"config":414},{"src":415,"dataGaName":412,"dataGaLocation":407},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1758203875/gs4c8p8opsgvflgkswz9.svg",{"text":417,"config":418},"Get Started",{"href":419,"dataGaName":420,"dataGaLocation":407},"https://gitlab.com/-/trial_registrations/new?glm_source=about.gitlab.com/get-started/","get started",{"freeTrial":422,"mobileIcon":426,"desktopIcon":428},{"text":423,"config":424},"Learn more about GitLab Duo",{"href":80,"dataGaName":425,"dataGaLocation":407},"gitlab duo",{"altText":409,"config":427},{"src":411,"dataGaName":412,"dataGaLocation":407},{"altText":409,"config":429},{"src":415,"dataGaName":412,"dataGaLocation":407},{"button":431,"mobileIcon":436,"desktopIcon":438},{"text":432,"config":433},"/switch",{"href":434,"dataGaName":435,"dataGaLocation":407},"#contact","switch",{"altText":409,"config":437},{"src":411,"dataGaName":412,"dataGaLocation":407},{"altText":409,"config":439},{"src":440,"dataGaName":412,"dataGaLocation":407},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1773335277/ohhpiuoxoldryzrnhfrh.png",{"freeTrial":442,"mobileIcon":447,"desktopIcon":449},{"text":443,"config":444},"Back to pricing",{"href":188,"dataGaName":445,"dataGaLocation":407,"icon":446},"back to pricing","GoBack",{"altText":409,"config":448},{"src":411,"dataGaName":412,"dataGaLocation":407},{"altText":409,"config":450},{"src":415,"dataGaName":412,"dataGaLocation":407},{"title":452,"button":453,"config":458},"See how agentic AI transforms software delivery",{"text":454,"config":455},"Watch GitLab Transcend now",{"href":456,"dataGaName":457,"dataGaLocation":46},"/events/transcend/virtual/","transcend event",{"layout":459,"icon":460,"disabled":27},"release","AiStar",{"data":462},{"text":463,"source":464,"edit":470,"contribute":475,"config":480,"items":485,"minimal":690},"Git is a trademark of Software Freedom Conservancy and our use of 'GitLab' is under license",{"text":465,"config":466},"View page source",{"href":467,"dataGaName":468,"dataGaLocation":469},"https://gitlab.com/gitlab-com/marketing/digital-experience/about-gitlab-com/","page source","footer",{"text":471,"config":472},"Edit this page",{"href":473,"dataGaName":474,"dataGaLocation":469},"https://gitlab.com/gitlab-com/marketing/digital-experience/about-gitlab-com/-/blob/main/content/","web ide",{"text":476,"config":477},"Please contribute",{"href":478,"dataGaName":479,"dataGaLocation":469},"https://gitlab.com/gitlab-com/marketing/digital-experience/about-gitlab-com/-/blob/main/CONTRIBUTING.md/","please contribute",{"twitter":481,"facebook":482,"youtube":483,"linkedin":484},"https://twitter.com/gitlab","https://www.facebook.com/gitlab","https://www.youtube.com/channel/UCnMGQ8QHMAnVIsI3xJrihhg","https://www.linkedin.com/company/gitlab-com",[486,533,585,629,656],{"title":186,"links":487,"subMenu":502},[488,492,497],{"text":489,"config":490},"View plans",{"href":188,"dataGaName":491,"dataGaLocation":469},"view plans",{"text":493,"config":494},"Why Premium?",{"href":495,"dataGaName":496,"dataGaLocation":469},"/pricing/premium/","why premium",{"text":498,"config":499},"Why Ultimate?",{"href":500,"dataGaName":501,"dataGaLocation":469},"/pricing/ultimate/","why ultimate",[503],{"title":504,"links":505},"Contact Us",[506,509,511,513,518,523,528],{"text":507,"config":508},"Contact sales",{"href":55,"dataGaName":56,"dataGaLocation":469},{"text":362,"config":510},{"href":364,"dataGaName":365,"dataGaLocation":469},{"text":367,"config":512},{"href":369,"dataGaName":370,"dataGaLocation":469},{"text":514,"config":515},"Status",{"href":516,"dataGaName":517,"dataGaLocation":469},"https://status.gitlab.com/","status",{"text":519,"config":520},"Terms of use",{"href":521,"dataGaName":522,"dataGaLocation":469},"/terms/","terms of use",{"text":524,"config":525},"Privacy statement",{"href":526,"dataGaName":527,"dataGaLocation":469},"/privacy/","privacy statement",{"text":529,"config":530},"Cookie preferences",{"dataGaName":531,"dataGaLocation":469,"id":532,"isOneTrustButton":27},"cookie preferences","ot-sdk-btn",{"title":91,"links":534,"subMenu":543},[535,539],{"text":536,"config":537},"DevSecOps platform",{"href":73,"dataGaName":538,"dataGaLocation":469},"devsecops platform",{"text":540,"config":541},"AI-Assisted Development",{"href":80,"dataGaName":542,"dataGaLocation":469},"ai-assisted development",[544],{"title":545,"links":546},"Topics",[547,552,557,562,567,570,575,580],{"text":548,"config":549},"CICD",{"href":550,"dataGaName":551,"dataGaLocation":469},"/topics/ci-cd/","cicd",{"text":553,"config":554},"GitOps",{"href":555,"dataGaName":556,"dataGaLocation":469},"/topics/gitops/","gitops",{"text":558,"config":559},"DevOps",{"href":560,"dataGaName":561,"dataGaLocation":469},"/topics/devops/","devops",{"text":563,"config":564},"Version Control",{"href":565,"dataGaName":566,"dataGaLocation":469},"/topics/version-control/","version control",{"text":23,"config":568},{"href":569,"dataGaName":37,"dataGaLocation":469},"/topics/devsecops/",{"text":571,"config":572},"Cloud Native",{"href":573,"dataGaName":574,"dataGaLocation":469},"/topics/cloud-native/","cloud native",{"text":576,"config":577},"AI for Coding",{"href":578,"dataGaName":579,"dataGaLocation":469},"/topics/devops/ai-for-coding/","ai for coding",{"text":581,"config":582},"Agentic AI",{"href":583,"dataGaName":584,"dataGaLocation":469},"/topics/agentic-ai/","agentic ai",{"title":586,"links":587},"Solutions",[588,590,592,597,601,604,608,611,613,616,619,624],{"text":133,"config":589},{"href":128,"dataGaName":133,"dataGaLocation":469},{"text":122,"config":591},{"href":105,"dataGaName":106,"dataGaLocation":469},{"text":593,"config":594},"Agile development",{"href":595,"dataGaName":596,"dataGaLocation":469},"/solutions/agile-delivery/","agile delivery",{"text":598,"config":599},"SCM",{"href":118,"dataGaName":600,"dataGaLocation":469},"source code management",{"text":548,"config":602},{"href":111,"dataGaName":603,"dataGaLocation":469},"continuous integration & delivery",{"text":605,"config":606},"Value stream management",{"href":161,"dataGaName":607,"dataGaLocation":469},"value stream management",{"text":553,"config":609},{"href":610,"dataGaName":556,"dataGaLocation":469},"/solutions/gitops/",{"text":171,"config":612},{"href":173,"dataGaName":174,"dataGaLocation":469},{"text":614,"config":615},"Small business",{"href":178,"dataGaName":179,"dataGaLocation":469},{"text":617,"config":618},"Public sector",{"href":183,"dataGaName":184,"dataGaLocation":469},{"text":620,"config":621},"Education",{"href":622,"dataGaName":623,"dataGaLocation":469},"/solutions/education/","education",{"text":625,"config":626},"Financial services",{"href":627,"dataGaName":628,"dataGaLocation":469},"/solutions/finance/","financial services",{"title":191,"links":630},[631,633,635,637,640,642,644,646,648,650,652,654],{"text":203,"config":632},{"href":205,"dataGaName":206,"dataGaLocation":469},{"text":208,"config":634},{"href":210,"dataGaName":211,"dataGaLocation":469},{"text":213,"config":636},{"href":215,"dataGaName":216,"dataGaLocation":469},{"text":218,"config":638},{"href":220,"dataGaName":639,"dataGaLocation":469},"docs",{"text":241,"config":641},{"href":243,"dataGaName":244,"dataGaLocation":469},{"text":236,"config":643},{"href":238,"dataGaName":239,"dataGaLocation":469},{"text":251,"config":645},{"href":253,"dataGaName":254,"dataGaLocation":469},{"text":259,"config":647},{"href":261,"dataGaName":262,"dataGaLocation":469},{"text":264,"config":649},{"href":266,"dataGaName":267,"dataGaLocation":469},{"text":269,"config":651},{"href":271,"dataGaName":272,"dataGaLocation":469},{"text":274,"config":653},{"href":276,"dataGaName":277,"dataGaLocation":469},{"text":279,"config":655},{"href":281,"dataGaName":282,"dataGaLocation":469},{"title":293,"links":657},[658,660,662,664,666,668,670,674,679,681,683,685],{"text":300,"config":659},{"href":302,"dataGaName":295,"dataGaLocation":469},{"text":305,"config":661},{"href":307,"dataGaName":308,"dataGaLocation":469},{"text":313,"config":663},{"href":315,"dataGaName":316,"dataGaLocation":469},{"text":318,"config":665},{"href":320,"dataGaName":321,"dataGaLocation":469},{"text":323,"config":667},{"href":325,"dataGaName":326,"dataGaLocation":469},{"text":328,"config":669},{"href":330,"dataGaName":331,"dataGaLocation":469},{"text":671,"config":672},"Sustainability",{"href":673,"dataGaName":671,"dataGaLocation":469},"/sustainability/",{"text":675,"config":676},"Diversity, inclusion and belonging (DIB)",{"href":677,"dataGaName":678,"dataGaLocation":469},"/diversity-inclusion-belonging/","Diversity, inclusion and belonging",{"text":333,"config":680},{"href":335,"dataGaName":336,"dataGaLocation":469},{"text":343,"config":682},{"href":345,"dataGaName":346,"dataGaLocation":469},{"text":348,"config":684},{"href":350,"dataGaName":351,"dataGaLocation":469},{"text":686,"config":687},"Modern Slavery Transparency Statement",{"href":688,"dataGaName":689,"dataGaLocation":469},"https://handbook.gitlab.com/handbook/legal/modern-slavery-act-transparency-statement/","modern slavery transparency statement",{"items":691},[692,695,698],{"text":693,"config":694},"Terms",{"href":521,"dataGaName":522,"dataGaLocation":469},{"text":696,"config":697},"Cookies",{"dataGaName":531,"dataGaLocation":469,"id":532,"isOneTrustButton":27},{"text":699,"config":700},"Privacy",{"href":526,"dataGaName":527,"dataGaLocation":469},[702],{"id":703,"title":9,"body":25,"config":704,"content":706,"description":25,"extension":24,"meta":710,"navigation":27,"path":711,"seo":712,"stem":713,"__hash__":714},"blogAuthors/en-us/blog/authors/itzik-gan-baruch.yml",{"template":705},"BlogAuthor",{"name":9,"config":707},{"headshot":708,"ctfId":709},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1749658921/Blog/Author%20Headshots/iganbaruch-headshot.jpg","iganbaruch",{},"/en-us/blog/authors/itzik-gan-baruch",{},"en-us/blog/authors/itzik-gan-baruch","bz9VMiTQ1ixvnoxUFk0jiUcnLG3oQsymgXNCqyRqfsk",[716,729,742],{"content":717,"config":727},{"title":718,"description":719,"authors":720,"heroImage":722,"date":723,"body":724,"category":11,"tags":725},"GitLab and Anthropic: Governed AI for enterprise development","GitLab deepens its Anthropic Claude integration, bringing governed AI, access to new models, and cloud flexibility to enterprise software development.",[721],"Stuart Moncada","https://res.cloudinary.com/about-gitlab-com/image/upload/v1776457632/llddiylsgwuze0u1rjks.png","2026-04-28","For enterprise and public sector leaders, the tension is familiar: Software teams need to move faster with AI, while security, compliance, and regulatory expectations only get more stringent. GitLab deepens its Anthropic Claude integration so organizations get access to newly released Claude models inside GitLab’s intelligent orchestration platform where governance, compliance, and auditability already run.\n\nClaude powers capabilities across GitLab Duo Agent Platform as the default model out of the box, across a variety of use cases from code generation and review to agentic chat and vulnerability resolution. If you've used GitLab Duo, you've already experienced how Duo agents automate workflows across the entire software development lifecycle (SDLC).\n\nThis accelerates the integration of Claude’s capabilities into GitLab, broadens how enterprises can deploy them, and reinforces what makes GitLab fundamentally different as a platform for software development and engineering: governance, compliance, and auditability built into every AI interaction.\n\n> \"GitLab Duo has accelerated how our teams plan, build, and ship software. The combination of Anthropic's Claude and GitLab's platform means we're getting more capable AI without changing how we work or how it is governed.\"\n>\n> – Mans Booijink, Operations Manager, Cube\n\n## The real differentiator: Governed AI\n\nWith GitLab, governance controls and auditing are built into the SDLC. When Claude suggests a code change through the GitLab Duo Agent Platform, that suggestion flows through the same merge request process, the same approval rules, the same security scanning, and the same audit trail as every other change. AI doesn't get a shortcut around your controls. It operates within them.\n\nAs GitLab moves deeper into agentic software development, where AI autonomously handles well-defined tasks, the governance layer becomes more important. An AI agent that can open a merge request, help resolve a vulnerability, or refactor a service needs to be auditable, attributable, and subject to the same policy enforcement as a human developer. That requirement is an architectural decision GitLab made from the start, and one that grows more consequential as AI agents take on broader responsibilities.\n\n## Enterprise deployment flexibility\n\nThis also expands how organizations access the latest Claude models through GitLab. Claude is available within GitLab through Google Cloud's Vertex AI and Amazon Bedrock, which means enterprises can route AI workloads through the hyperscaler commitments and cloud governance frameworks they already have in place. No separate vendor contract. No new data residency questions. Your existing Google Cloud or AWS relationship is the on-ramp. \n\nGitLab is now also available in the [Claude Marketplace](https://claude.com/platform/marketplace), allowing customers to purchase GitLab Credits and apply them toward existing Anthropic spending commitments – consolidating AI spend and simplifying how teams discover and procure GitLab alongside their Anthropic investments.\n\n## Advancing an agentic future\n\nGitLab's vision for agentic software development, where AI handles defined tasks autonomously across planning, coding, testing, securing, and deploying, requires models with strong reasoning, reliability, and safety characteristics. It also requires a platform where those autonomous actions are fully governed.\n\nAgentic workflows demand models with strong reasoning, reliability, and safety characteristics, criteria that guide how GitLab selects and integrates AI model partners. And GitLab's governance framework helps ensure that as AI agents assume more advanced development work, enterprises maintain full visibility and control over what those agents do, when they do it, and how changes are tracked.\n\n## What this means for GitLab customers\n\nIf you're already using GitLab Duo Agent Platform, you'll get access to Claude models and deeper AI assistance across your software development lifecycle, all within the governance framework you already rely on.\n\nIf you're evaluating AI-powered software development platforms, you shouldn't have to choose between advanced AI capabilities and enterprise control. This strategic integration is built to deliver both.\n\n> Want to learn more about GitLab Duo Agent Platform? [Get a demo or start a free trial today](https://about.gitlab.com/gitlab-duo-agent-platform/).",[22,726,282],"product",{"featured":27,"template":15,"slug":728},"gitlab-and-anthropic-governed-ai-for-enterprise-development",{"content":730,"config":740},{"title":731,"description":732,"authors":733,"heroImage":735,"date":736,"body":737,"category":11,"tags":738},"Give your AI agent direct, structured GitLab access with glab CLI","The GitLab CLI (glab) provides AI agents structured, reliable access to projects via the MCP, eliminating friction. This tutorial shows how you can speed up code review and issue triage.",[734],"Kai Armstrong","https://res.cloudinary.com/about-gitlab-com/image/upload/v1776347152/unw3mzatkd5xyfbzcnni.png","2026-04-27","\nWhen teams use GitLab Duo, Claude, Cursor, and other AI assistants, more of the development workflow runs through an AI agent acting on your behalf — reading issues, reviewing merge requests, running pipelines, and helping you ship faster. Most developers are already using the GitLab CLI (`glab`) from the terminal to interact with GitLab. Combining the two is a natural next step.\n\n\nThe problem is that without the right tools, AI agents are essentially guessing when it comes to your GitLab projects. They might hallucinate the details of an issue they've never seen, summarize a merge request based on stale training data rather than its actual state, or require you to manually copy context from a browser tab and paste it into a chat window just to get started. Every one of those workarounds is friction: it slows you down, introduces the possibility of error, and puts a hard ceiling on what your agent can actually do on your behalf. `glab` changes that by giving agents a direct, reliable interface to your projects.\n\n\nWith `glab`, your agent fetches what it needs directly from GitLab, acts on it, and reports back — so you spend less time relaying information and more time on the work that matters.\n\n\nIn this tutorial, you'll learn how to use `glab` to give AI agents structured, reliable access to your GitLab projects. You'll also discover how that unlocks a faster, more capable development workflow.\n\n\n## How to connect your AI agent to GitLab through MCP\n\n\nThe most direct way to supercharge your AI workflow is to give your AI agent native access to `glab` through Model Context Protocol ([MCP](https://about.gitlab.com/topics/ai/model-context-protocol/)).\n\n\n MCP is an open standard that lets AI tools discover and use external capabilities at runtime. Once connected, your AI assistant can read issues, comment on merge requests, check pipeline status, and write back to GitLab, all without copying anything from the UI or writing a single API call yourself.\n\n\n To get started, run:\n\n\n ```shell\n # Start the glab MCP server\n glab mcp serve\n ```\n\n\n Once your MCP client is configured, your AI can answer questions like *\"What's the status of my open MRs?\"* or *\"Are there any failing pipelines on main?\"* by querying GitLab directly, not scraping the web UI, not relying on stale training data. See the [full setup docs](https://docs.gitlab.com/cli/) for configuration steps for Claude Code, Cursor, and other editors.\n\n\n One detail worth knowing: `glab` automatically adds `--output json` when invoked through MCP, for any command that supports it. Your agent gets clean, structured data without you needing to think about output formats. And because `glab` uses the official MCP SDK, it stays compatible as the\n protocol evolves.\n\n\n We've also been deliberate about *which* commands are exposed through MCP. Commands that require interactive terminal input are intentionally\n excluded, so your agent never gets stuck waiting for input that will never come. What's exposed is what actually works reliably in an agent context.\n\n\n ## Let your AI participate in code review\n\n\n Most developers have a backlog of MRs waiting for review. It's one of the most time-consuming parts of the job and one of the best places to put\n AI to work. With `glab`, your agent doesn't just observe your review queue, it can work through it with you.\n\n\n ### See exactly what still needs addressing\n\n\n Start with this:\n\n\n ```shell\n glab mr view 2677 --comments --unresolved --output json\n ```\n\n\n This input returns the full MR: metadata, description, and every\n unresolved discussion, as a single structured JSON payload. Hand that to\n your AI and it has everything it needs: which threads are open, what the\n reviewer asked for, and in what context. No tab-switching, no copy-pasting\n individual comments.\n\n\n \n ```json\n {\n   \"id\": 2677,\n   \"title\": \"feat: add OAuth2 support\",\n   \"state\": \"opened\",\n   \"author\": { \"username\": \"jdwick\" },\n   \"labels\": [\"backend\", \"needs-review\"],\n   \"blocking_discussions_resolved\": false,\n   \"discussions\": [\n     {\n       \"id\": \"3107030349\",\n       \"resolved\": false,\n       \"notes\": [\n         {\n           \"author\": { \"username\": \"dmurphy\" },\n           \"body\": \"This error handling will swallow panics — consider wrapping with recover()\",\n           \"created_at\": \"2026-03-14T09:23:11.000Z\"\n         }\n       ]\n     },\n     {\n       \"id\": \"3107030412\",\n       \"resolved\": false,\n       \"notes\": [\n         {\n           \"author\": { \"username\": \"sreeves\" },\n           \"body\": \"Token refresh logic needs a test for the expired token case\",\n           \"created_at\": \"2026-03-14T10:05:44.000Z\"\n         }\n       ]\n     }\n   ]\n }\n ```\n\n\n Instead of reading through every thread yourself, you ask your agent  *\"what do I still need to fix in MR 2677?\"* and get back a prioritized summary with suggested changes. This all happens from a single command.\n\n\n ### Close the loop programmatically\n\n\n Once your AI has helped you address the feedback, it can resolve\n discussions:\n\n\n ```shell\n # List all discussions — structured, ready for the agent to process\n glab mr note list 456 --output json\n\n # Resolve a discussion once the feedback is addressed\n glab mr note resolve 456 3107030349\n\n # Reopen if something needs another look\n glab mr note reopen 456 3107030349\n ```\n\n\n\n ```json\n [\n   {\n     \"id\": 3107030349,\n     \"body\": \"This error handling will swallow panics — consider wrapping with recover()\",\n     \"author\": { \"username\": \"dmurphy\" },\n     \"resolved\": false,\n     \"resolvable\": true\n   },\n   {\n     \"id\": 3107030412,\n     \"body\": \"Token refresh logic needs a test for the expired token case\",\n     \"author\": { \"username\": \"sreeves\" },\n     \"resolved\": false,\n     \"resolvable\": true\n   }\n ]\n ```\n\n\n\n Note IDs are visible directly in the GitLab UI and API, no extra lookup needed. Your agent can work through the full list, verify each fix, and\n resolve as it goes.\n\n\n ## Talk to your AI about your code more effectively\n\n\n Even if you're not running an MCP server, there's a simpler shift that makes a huge difference: using `glab` to feed your AI better information.\n\n\n Think about the last time you asked an AI assistant to help triage issues or debug a failing pipeline. You probably copied some text from the GitLab UI and pasted it into the chat. Here's what your agent is actually\n working with when you do that:\n\n\n ```text\n open issues: 12 • milestone: 17.10 • label: bug, needs-triage ...\n ```\n\n\n Compare that to what it gets with `glab`:\n\n\n \n ```json\n [\n   {\n     \"iid\": 902,\n     \"title\": \"Pipeline fails on merge to main\",\n     \"labels\": [\"bug\", \"needs-triage\"],\n     \"milestone\": { \"title\": \"17.10\" },\n     \"assignees\": []\n   },\n   ...\n ]\n ```\n\n\n Structured, typed, complete; no ambiguity, no parsing guesswork. That's the difference between an agent that can act and one that has to ask\n follow-up questions.\n\n\n If you're using the MCP server, you get this automatically: `glab` adds `--output json` for any command that supports it. If you're working directly\n from the terminal, just add the flag yourself:\n\n\n ```shell\n # Pull open issues for triage\n glab issue list --label \"needs-triage\" --output json\n\n # Check pipeline status\n glab ci status --output json\n\n # Get full MR details\n glab mr view 456 --output json\n ```\n\n\n We've significantly expanded JSON output support in recent releases. It now covers CI status, milestones, labels, releases, schedules, cluster agents, work items, MR approvers, repo contributors, and more. If `glab` can\n retrieve it, your AI can consume it cleanly.\n\n\n ### A real workflow\n\n\n ```shell\n $ glab issue list --label \"needs-triage\" --milestone \"17.10\"\n --output json\n ```\n\n\n ```text\n Agent: I found 2 unassigned bugs in the 17.10 milestone that need triage:\n 1. #902 — Pipeline fails on merge to main (opened 5 days ago)\n 2. #903 — Auth token not refreshing on expiry (opened 4 days ago)\n Both are unassigned. Want me to draft triage notes and suggest assignees based on recent commit history?\n ```\n\n\n ## Your agent is never limited to built-in commands\n\n\n `glab`'s first-class commands cover the most common workflows, but your agent is never limited to them. Through `glab api`, it has authenticated access to the full GitLab REST and GraphQL API surface, using the same session, with no extra credentials or configuration required.\n\n\n This is a meaningful differentiator. Most CLI tools stop at what their commands expose. With `glab`, if GitLab's API supports it, your agent can do it. It's always working from a trusted, authenticated context.\n\n\n A practical example: fetching just the list of changed files in an MR before deciding which diffs to pull in full:\n\n\n ```shell\n # Get changed file paths — lightweight, no diff content yet\n glab api \"/projects/$CI_PROJECT_ID/merge_requests/$CI_MERGE_REQUEST_IID/diffs?per_page=100\" \\\n | jq '.[].new_path'\n\n# Then fetch only the specific file your agent needs\nglab api \"/projects/$CI_PROJECT_ID/merge_requests/$CI_MERGE_REQUEST_IID/diffs?per_page=100\" \\\n| jq '.[] | select(.new_path == \"path/to/file.go\")'\n ```\n\n\n ```text\n \"internal/auth/token.go\"\n \"internal/auth/token_test.go\"\n \"internal/oauth/refresh.go\"\n ```\n\n\n For anything the REST API doesn't cover (epics, certain work item queries, complex cross-project data),  `glab api graphql` gives you the full\n GraphQL interface:\n\n\n ```shell\n   glab api graphql -f query='\n {\n   project(fullPath: \"gitlab-org/gitlab\") {\n     mergeRequest(iid: \"12345\") {\n       title\n       reviewers { nodes { username } }\n     }\n   }\n }'\n ```\n\n ```json\n{\n   \"data\": {\n     \"project\": {\n       \"mergeRequest\": {\n         \"title\": \"feat: add OAuth2 support\",\n         \"reviewers\": {\n           \"nodes\": [\n             { \"username\": \"dmurphy\" },\n             { \"username\": \"sreeves\" }\n           ]\n         }\n       }\n     }\n   }\n }\n\n ```\n\n\n Your agent has a single, authenticated entry point to everything GitLab exposes without the token juggling, separate API clients, or configuration\n overhead.\n\n\n ## What's coming and your feedback\n\n\n Two improvements we're actively working on will make `glab` even more useful for agent workflows:\n\n\n **Agent-aware help text.** Today, `--help` output is written for humansvat a terminal. We're updating it to surface the non-interactive alternative\n for every interactive command, flag which commands support `--output json`, and generally make help a useful resource for agents discovering\n capabilities at runtime — not just humans.\n\n\n **Better machine-readable errors.** When something goes wrong today, agents get the same human-readable error messages as terminal users. We're\n changing that so errors in JSON mode return structured output, giving your agent the information it needs to handle failures gracefully, retry intelligently, or surface the right context back to you.\n\n\n Both of these are in active development. If you're already using `glab` with an AI tool, you're exactly the audience we want feedback from.\n\n\n * **What friction are you hitting?** Commands that don't behave well in agent contexts, error messages that aren't actionable, gaps in JSON output\n coverage. We want to know.\n\n * **What workflows have you unlocked?** Real usage patterns help us prioritize what to build next.\n\n\n Join the discussion in [our feedback issue](https://gitlab.com/gitlab-org/cli/-/issues/8177) — that's where we're shaping the roadmap for agent-friendliness, and where your input will have the most direct impact. If you've found a specific gap, [open an issue](https://gitlab.com/gitlab-org/cli/-/issues/new). If you've got a fix in mind, contributions are welcome. Visit [CONTRIBUTING.md](https://gitlab.com/gitlab-org/cli/-/blob/main/CONTRIBUTING.md) to get started.\n\n\n The GitLab CLI has always been about giving developers more control over their workflow. As AI becomes a bigger part of how we all work, that means making `glab` the best possible interface between your AI tools and your GitLab projects. We're just getting started and we'd love to build the next part with you.\n",[22,726,739],"tutorial",{"featured":27,"template":15,"slug":741},"give-your-ai-agent-direct-structured-gitlab-access-with-glab-cli",{"content":743,"config":751},{"title":744,"description":745,"authors":746,"heroImage":735,"date":748,"body":749,"category":11,"tags":750},"GitHub Copilot's new policy for AI training is a governance wake-up call","Learn what GitHub's Copilot policy change means for regulated industries, and why GitLab's commitment to customer data privacy matters.",[747],"Allie Holland","2026-04-20","GitHub recently [announced](https://github.blog/news-insights/company-news/updates-to-github-copilot-interaction-data-usage-policy/) a significant change to how it handles data from Copilot users. Starting April 24, 2026, interaction data from Copilot Free, Pro, and Pro+ users, including inputs, outputs, code snippets, and associated context, will be used to train AI models by default, unless users actively opt out. Copilot Business and Enterprise customers are exempt under existing contract terms.\n\nFor organizations in regulated industries, including finance, healthcare, defense, and public sector, the policy shift raises questions that go beyond individual developer preferences. It forces a harder look at a question that engineering and security leaders should be asking every AI vendor in their stack: Do you train on our code? \n\nGitLab's answer is no. GitLab does not train AI models on customer code at any tier, and AI vendors are contractually prohibited from using customer inputs or outputs for their own purposes. The [GitLab AI Transparency Center](https://about.gitlab.com/ai-transparency-center/) makes that commitment auditable: a single location documenting which models power which features, how data is handled, subprocessor relationships, and data retention periods. The GitLab AI Transparency Center also lists the compliance status of each feature, including confirmation that GitLab's current AI features do not qualify as high-risk systems under the EU AI Act. It's a standard GitLab CEO Bill Staples has consistently [reiterated](https://www.linkedin.com/posts/williamstaples_gitlab-1810-agentic-ai-now-open-to-even-activity-7443280763715985408-aHxf?utm_source=share&utm_medium=member_desktop&rcm=ACoAABsu7EUBcb_a1-JHKS9RC0B5rf8Ye-5XM60) and one reflected in GitLab's mission and [Trust Center](https://trust.gitlab.com/).\n\n## What the policy change actually means\n\nGitHub's announcement also specifies that the data may be shared with GitHub affiliates, including Microsoft, for AI development purposes.\n\nA policy change of this nature forces organizations to re-examine their AI governance posture, audit their Copilot license tiers, and confirm that the right controls are configured across their teams.\n\n## Why AI governance matters in regulated environments\n\nSource code is often among an organization's most sensitive intellectual property. It may contain references to internal systems, reflect proprietary business logic, or touch data flows governed by strict retention and access policies. When that code passes through an AI assistant, questions about training data usage, model vendor relationships, and data residency become compliance concerns.\n\nThe exposure is particularly acute for financial services firms that have invested in proprietary algorithms, fraud detection logic, credit risk models, underwriting rules, trading strategies. When AI tooling processes that code and uses it to train models serving competitors, vendor data practices become an IP concern.\n\nFinancial institutions operating under [the Federal Reserve's Supervisory Guidance on Model Risk Management (SR 11-7) and the](https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm) [Digital Operational Resilience Act (DORA)](https://eur-lex.europa.eu/eli/reg/2022/2554/oj/eng) are required to maintain documented, auditable oversight of third-party technology providers, including understanding how those providers handle data. Third-party AI tools used in development workflows increasingly fall within the scope of model risk oversight, and material changes to vendor data practices require updated documentation. \n\nIn the public sector, [the National Institute of Standards and Technology Special Publication 800-53 (NIST 800-53)](https://csrc.nist.gov/publications/detail/sp/800-53/rev-5/final) and the [Federal Information Security Modernization Act (FISMA)](https://www.cisa.gov/topics/cyber-threats-and-advisories/federal-information-security-modernization-act) establish that sensitive or classified code must never leave a controlled boundary. For U.S. Department of Defense and intelligence community environments in particular, a vendor's default data posture is an operational concern. In healthcare, [the Health Insurance Portability and Accountability Act (HIPAA)](https://www.hhs.gov/hipaa/index.html) governs how patient-adjacent data is handled by third parties, and development environments that touch clinical systems increasingly fall within that scope.\n\nAcross all of these contexts, the common thread is the same: A vendor policy that changes data usage defaults, requires individual opt-out, and offers different protections depending on account tier introduces exactly the kind of uncontrolled variable that compliance teams cannot afford.\n\n## What regulated industries actually need from AI vendors\n\nRegulated organizations have largely moved past debating whether to adopt AI in development workflows. The focus now is on doing so in a way they can defend to regulators, boards, and customers. That shift has surfaced a consistent set of requirements regardless of sector.\n\n**Contractual certainty.** Regulated firms need to know, with specificity, what happens to their data. A clear, documented, unconditional commitment is what's required, not something that varies by plan or requires action before a deadline.\n\n**Auditability.** Model risk management frameworks require organizations to understand and validate the AI systems they deploy, including the training data behind those models and the third parties involved in their development. Vendors who cannot answer these questions create documentation risk for the organizations relying on them.\n\n**Separation from vendor incentives.** When an AI vendor trains models on customer usage data, code and workflows become inputs to a system that also serves competitors. For institutions with proprietary trading logic, underwriting models, or fraud detection systems, that's a genuine IP exposure.\n\n## GitLab's position on AI data governance\n\nGitLab does not use customer code to train AI models. This commitment applies at every tier, and AI vendors are contractually prohibited from using inputs or outputs associated with GitLab customers for their own purposes.\n\nThis is a deliberate architectural and policy choice, not a feature of a particular pricing tier. As GitLab's [post on enterprise independence](https://about.gitlab.com/blog/why-enterprise-independence-matters-more-than-ever-in-devsecops/) notes, data governance has become \"an increasingly critical factor in enterprise technology decisions, driven by a complex web of national and regional data protection laws and growing concern about control over sensitive intellectual property.\"\n\nGitLab is also cloud-neutral and model-neutral while supporting self-hosted deployments, not commercially tied to any single cloud provider or large language model (LLM). That i[ndependence matters](https://about.gitlab.com/blog/why-enterprise-independence-matters-more-than-ever-in-devsecops/) for regulated organizations evaluating vendor concentration risk. The [AI Continuity Plan](https://handbook.gitlab.com/handbook/product/ai/continuity-plan/) documents how vendor changes are managed, including material changes to how AI vendors treat customer data, a direct response to the governance requirements under frameworks like [DORA](https://handbook.gitlab.com/handbook/legal/dora/). \n\n## The governance gap AI teams need to close\n\nGitHub's policy update is a reminder that for organizations in regulated industries, understanding exactly how an AI tool handles data is a prerequisite for using it at all. That means asking vendors for clear, documented answers: Is our data used for model training? Who are your AI model subprocessors? What happens if a vendor changes its data practices? Can we deploy in a way that keeps all AI processing within our own infrastructure? What indemnification do you offer for AI-generated output?\n\nVendors who can answer those questions clearly, and document those answers in an auditable form, are vendors you can build on. **Those who cannot will create compliance debt every time they ship a policy update.** And when a vendor can change its data practices with 30 days notice, that's not a partnership built for regulated industries. That's a liability.\n\n> Learn more about GitLab's approach to AI governance at the [GitLab AI Transparency Center](https://about.gitlab.com/ai-transparency-center/).",[22,726],{"featured":14,"template":15,"slug":752},"github-copilots-new-policy-for-ai-training-is-a-governance-wake-up-call",{"promotions":754},[755,768,779,791],{"id":756,"categories":757,"header":758,"text":759,"button":760,"image":765},"ai-modernization",[11],"Is AI achieving its promise at scale?","Quiz will take 5 minutes or less",{"text":761,"config":762},"Get your AI maturity score",{"href":763,"dataGaName":764,"dataGaLocation":244},"/assessments/ai-modernization-assessment/","modernization assessment",{"config":766},{"src":767},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1772138786/qix0m7kwnd8x2fh1zq49.png",{"id":769,"categories":770,"header":771,"text":759,"button":772,"image":776},"devops-modernization",[726,37],"Are you just managing tools or shipping innovation?",{"text":773,"config":774},"Get your DevOps maturity score",{"href":775,"dataGaName":764,"dataGaLocation":244},"/assessments/devops-modernization-assessment/",{"config":777},{"src":778},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1772138785/eg818fmakweyuznttgid.png",{"id":780,"categories":781,"header":783,"text":759,"button":784,"image":788},"security-modernization",[782],"security","Are you trading speed for security?",{"text":785,"config":786},"Get your security maturity score",{"href":787,"dataGaName":764,"dataGaLocation":244},"/assessments/security-modernization-assessment/",{"config":789},{"src":790},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1772138786/p4pbqd9nnjejg5ds6mdk.png",{"id":792,"paths":793,"header":796,"text":797,"button":798,"image":803},"github-azure-migration",[794,795],"migration-from-azure-devops-to-gitlab","integrating-azure-devops-scm-and-gitlab","Is your team ready for GitHub's Azure move?","GitHub is already rebuilding around Azure. Find out what it means for you.",{"text":799,"config":800},"See how GitLab compares to GitHub",{"href":801,"dataGaName":802,"dataGaLocation":244},"/compare/gitlab-vs-github/github-azure-migration/","github azure migration",{"config":804},{"src":778},{"header":806,"blurb":807,"button":808,"secondaryButton":813},"Start building faster today","See what your team can do with the intelligent orchestration platform for DevSecOps.\n",{"text":809,"config":810},"Get your free trial",{"href":811,"dataGaName":51,"dataGaLocation":812},"https://gitlab.com/-/trial_registrations/new?glm_content=default-saas-trial&glm_source=about.gitlab.com/","feature",{"text":507,"config":814},{"href":55,"dataGaName":56,"dataGaLocation":812},1777493613627]