# GitHub Copilot

> Source: https://aiwiki.ai/wiki/github_copilot
> Updated: 2026-07-29
> Fact-checked: 2026-07-29
> Categories: AI Code Generation, AI Tools & Products, Developer Tools, Microsoft
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "GitHub Copilot." aiwiki.ai, 29 Jul 2026. https://aiwiki.ai/wiki/github_copilot
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

GitHub Copilot is a hosted [artificial intelligence](https://aiwiki.ai/wiki/artificial_intelligence) coding assistant developed by [GitHub](https://aiwiki.ai/wiki/github). It began as a code-completion extension built with [OpenAI](https://aiwiki.ai/wiki/openai) and has expanded into a multi-model product that offers inline suggestions, chat, code review, local and cloud-based coding agents, a command-line interface, and a standalone desktop application. GitHub announced a technical preview on June 29, 2021, and general availability for individual developers on June 21, 2022.[1][2]

The product is not a single model. GitHub operates the service and integrates models from several providers, with availability depending on the plan, client, feature, and organization policy. The supported-model catalog changes independently of the product name. For that reason, a current model should not be inferred from an older description of Copilot, and generated code should be treated as a proposed change that still requires human review, testing, and security analysis.[3][4]

## Product scope

GitHub uses "Copilot" for a family of related developer experiences. The original experience predicts code in an editor. Copilot Chat answers questions and proposes changes using conversational context. Agent mode can inspect a local workspace, edit files, invoke tools, and run commands with user-approved permissions. Copilot cloud agent works asynchronously in a GitHub-managed environment and can create or update a pull request. Copilot CLI provides an agent in a terminal, while the GitHub Copilot app provides a desktop interface. Copilot code review comments on selected changes in a pull request or editor.[3][5][6][7][8]

These surfaces have different trust boundaries. Editor completions operate in the developer's active editing context. Agent mode acts in a local development environment. The CLI can interact with files and commands within its granted scope. The cloud agent uses an ephemeral environment backed by GitHub Actions, commits to a branch, and works through a pull request. Product controls that apply to one surface do not necessarily apply to another.[5][6][9]

### Code completion and next edit suggestions

Inline code completion is Copilot's oldest feature. The client sends selected context to the service and displays a proposed continuation in the editor. A suggestion can be a short token sequence, a line, or a larger block. The developer may accept, modify, or reject it. GitHub documents particularly strong support for Python, JavaScript, TypeScript, Ruby, Go, C#, and C++, while also supporting many other languages with varying quality.[10]

Next edit suggestions predict the location and content of a likely subsequent edit after a developer changes code. GitHub introduced the feature in preview alongside agent mode in February 2025. Unlike a chat request, the feature is intended to keep a developer in the normal editing flow. Under the billing rules in effect from June 1, 2026, code completions and next edit suggestions are included in plans and do not consume GitHub AI Credits.[11][12]

### Chat and multi-file editing

Copilot Chat accepts natural-language questions about code, development tools, and selected repository context. Depending on the client, it can explain code, propose tests, suggest fixes, generate terminal commands, or prepare edits. GitHub made Copilot Chat generally available in December 2023. The exact context available to chat varies by client and by whether the user explicitly attaches a file, symbol, selection, repository, or other source.[13][3]

Multi-file editing developed from the earlier Copilot Edits experience. A developer describes a change, reviews proposed modifications across a working set, and can accept or reject individual edits. Agent mode goes further by deciding which files and tools may be relevant, executing a sequence of steps, and reacting to command or test results. Neither workflow makes the generated change intrinsically correct.[3][4]

### Local agent mode

Agent mode is the local, editor-based agentic workflow. It may search a repository, edit several files, call configured tools, run terminal commands, and iterate after errors. Its permissions and confirmation behavior depend on the client and user settings. Connections to external tools can include the [Model Context Protocol](https://aiwiki.ai/wiki/model_context_protocol), but an MCP server extends the agent's access and must be evaluated as part of the security boundary.[11][14]

Agent mode is distinct from cloud agent. Local agent mode acts in the developer's current workspace and can use locally available tools. Cloud agent runs on GitHub's infrastructure and communicates its work through a GitHub branch and pull request. A statement about the limits or data controls of one should not be generalized to the other.[5][9]

### Copilot cloud agent

GitHub introduced its asynchronous coding agent in May 2025 and made it generally available in September 2025. In April 2026, GitHub renamed and expanded the experience as Copilot cloud agent, adding research and planning tasks as well as code changes. A user can delegate work from supported GitHub interfaces, then review the resulting branch, session logs, and pull request.[15][16][17]

As documented at the evidence cutoff, cloud agent is available on paid Copilot plans. It runs in an ephemeral GitHub Actions environment, is limited to one repository and one branch per task, and can open exactly one pull request for each task. GitHub also documented a maximum session duration of 59 minutes. Cloud-agent use consumes AI Credits and GitHub Actions minutes. Content exclusions do not apply to this surface, which is an important distinction for organizations handling restricted code.[5][9]

### Copilot CLI and desktop app

The newer terminal-native GitHub Copilot CLI agent reached general availability in February 2026. It is distinct from the earlier gh-copilot extension, described as Copilot in the CLI, that reached general availability in March 2024. The newer agent is interactive and programmatic on Linux, macOS, Windows PowerShell, and Windows Subsystem for Linux. It can inspect files, propose edits, and request command execution. GitHub's documentation describes approval controls and trusted-directory behavior, but also warns that trust decisions are heuristic rather than a complete security guarantee.[6][18][55]

The GitHub Copilot app is a desktop agent interface for macOS, Windows, and Linux. GitHub made it available across Copilot plans in July 2026. The app can also be used with a user-supplied model key without a Copilot subscription, subject to the terms and data handling of the selected provider. The app and CLI are separate clients even when their workflows overlap.[7][19]

### Code review

Copilot code review analyzes selected changes and produces review comments. It can be requested for a pull request and, in supported clients, for local changes. GitHub describes its feedback as supplementary: it may miss problems, produce incorrect advice, or focus on maintainability rather than a particular security policy. A Copilot review is therefore not a substitute for accountable human review or dedicated static, dynamic, dependency, and secret-scanning tools.[8][4]

## How context is selected

Copilot output depends on the prompt, the model, and the context exposed by the client. Possible context includes the current file, selected text, open files, symbols, repository search results, conversation history, terminal output, and explicitly attached sources. Agentic clients may gather additional context by searching files or invoking tools.[3][20]

GitHub supports repository indexing to improve semantic retrieval. The index represents repository content for search and context selection; GitHub states that the indexed repository is not used to train its models merely because it was indexed. Indexing is not equivalent to sending the entire repository with every prompt, and access remains subject to repository permissions.[20]

Organizations on Copilot Business or Enterprise can configure content exclusions for supported surfaces. Excluded paths are intended to keep matching content out of certain Copilot suggestions and responses. GitHub documents material exceptions: content exclusions do not cover Copilot CLI, Copilot cloud agent, or agent mode in integrated development environments. An exclusion rule should therefore not be treated as a universal data-loss-prevention control.[9][21]

### Public code matching

GitHub's code-referencing system checks eligible suggestions for matches against an index of public GitHub code. Depending on the user's policy, a matching suggestion may be blocked or displayed with references to matching repositories and license information. The system compares a potential suggestion together with about 150 characters of surrounding code against the index. GitHub does not present 150 characters as a minimum match threshold. The system does not determine that a suggestion is legally safe, nor can a non-match prove originality.[22]

Repository administrators and users can choose policies that allow or block suggestions matching public code. The feature addresses a subset of verbatim or near-verbatim output risk, not the broader questions of training legality, software licensing, patents, or whether independently generated code implements a protected work. Users remain responsible for reviewing applicable licenses and obligations.[22][23]

## Development history

### Codex and the 2021 technical preview

GitHub announced Copilot as a Visual Studio Code technical preview on June 29, 2021. The launch described a system that used surrounding code and comments to suggest lines or functions, with the strongest early results reported for Python, JavaScript, TypeScript, Ruby, and Go. GitHub and OpenAI developed the original service together.[1]

The early product used a production version of [OpenAI Codex](https://aiwiki.ai/wiki/openai_codex). OpenAI's 2021 Codex paper described a GPT model fine-tuned on publicly available GitHub code and stated that a distinct production version powered GitHub Copilot. The paper evaluated Python program synthesis on HumanEval; its benchmark results describe the research models in that paper, not the accuracy of every Copilot suggestion or later Copilot model.[24]

GitHub made Copilot generally available to individuals on June 21, 2022, with historical pricing of $10 per month or $100 per year. It also offered free access to verified students and maintainers of qualifying open-source projects. The current plan catalog and billing rules have since changed.[2]

### Organization plans and Copilot X

GitHub made Copilot for Business available to invoiced GitHub Enterprise customers on December 7, 2022, then widened it to Free, Team, and GitHub Enterprise Cloud organizations on February 14, 2023. The organization offering added centralized license management, policy controls, and business-oriented terms. In March 2023, GitHub announced "Copilot X" as a product vision spanning chat, pull requests, documentation, and the command line. Copilot X was not a permanent rename of GitHub Copilot.[54][25][26]

Copilot Chat reached general availability in December 2023. Copilot Enterprise became generally available in February 2024, adding deeper GitHub.com and organization-context features. The details of both offerings have continued to evolve, so historical launch descriptions do not define the current feature set.[13][27]

### Multi-model support, free access, and agents

At GitHub Universe 2024, GitHub announced model choice, multi-file editing, code review, and broader client support. This marked a transition from a product identified with one OpenAI model to a platform that can route work to models from multiple providers. GitHub launched Copilot Free on December 18, 2024, with 2,000 code completions per month and a limited chat allowance.[28][29]

GitHub previewed next edit suggestions and agent mode in February 2025. It launched the asynchronous coding agent in May 2025 and made it generally available in September. The separate Copilot Workspace technical preview, which explored an issue-to-plan-to-code workflow in a browser, was shut down on May 30, 2025; it should not be listed as a current Copilot product.[11][15][16][30]

The product expanded beyond editors in 2026. The newer terminal-native Copilot CLI agent reached general availability in February, the coding agent became Copilot cloud agent in April, and the standalone Copilot app became available to all plans in July. These launches also increased the importance of distinguishing local tools, GitHub-hosted agents, and third-party model providers.[17][18][19][55]

## Clients, languages, and models

GitHub maintains a feature matrix rather than promising identical behavior in every editor. At the evidence cutoff, the matrix covered Visual Studio Code, Visual Studio, JetBrains IDEs, Eclipse, Xcode, and Neovim. Code completion was broadly available across those clients, while chat, agent mode, model selection, code review, and customization differed. Some entries were marked preview and can change without an article revision.[31]

The feature matrix, client release notes, and organization policy should be checked together when a specific capability matters. A feature available on GitHub.com or in the desktop app may not exist in an IDE extension with the same name.[31]

Copilot supports many programming languages, but support is not a binary compatibility claim. Suggestion quality depends on the language, framework, repository context, prompt, and task. GitHub's code-suggestion documentation identifies Python, JavaScript, TypeScript, Ruby, Go, C#, and C++ as languages for which the service may perform especially well.[10]

The current model catalog includes models from multiple providers and records separate availability for features and plans. Models can be added, deprecated, or restricted by an organization administrator. A static list quickly becomes inaccurate, so the official model catalog is the appropriate source for a deployment decision.[32]

## Plans and billing

The following prices were listed by GitHub at the evidence cutoff. They are base subscription prices before taxes, usage beyond included allowances, or other services. Verified-student eligibility and regional terms can add conditions.[33][34]

| Plan | Listed base price | Intended account type |
|---|---:|---|
| Free | $0 | Individual |
| Student | $0 for eligible verified students | Individual |
| Pro | $10 per month | Individual |
| Pro+ | $39 per month | Individual |
| Max | $100 per month | Individual |
| Business | $19 per granted seat per month | Organization |
| Enterprise | $39 per granted seat per month | Enterprise |

GitHub does not offer Copilot for GitHub Enterprise Server as a self-hosted product. Enterprise customers can use Copilot with supported GitHub cloud and client configurations, but that is different from running the Copilot service on premises.[33]

On June 1, 2026, GitHub replaced premium-request units with usage-based GitHub AI Credits. One GitHub AI Credit corresponds to $0.01 of metered usage. Plans include differing monthly credit allowances, and additional use can be governed by user or organization budgets. Input tokens, output tokens, cached tokens, models, and agent workflows can affect consumption. Code completion and next edit suggestions do not consume AI Credits under the announced rules.[12][34]

Code review consumes AI Credits and can also use GitHub Actions minutes. Cloud-agent sessions likewise use AI Credits and Actions resources. GitHub kept the listed base prices of Pro, Pro+, Business, and Enterprise unchanged during the 2026 transition, but the total cost of agent-heavy use can exceed the subscription price. Administrators can set budgets and monitor usage.[12][34]

## Administration and data handling

Business and Enterprise administrators can grant seats, control feature and model policies, manage budgets, and configure supported content exclusions. Access to a repository is still governed by GitHub permissions, while model and feature policy can be narrower than repository access. Third-party agents, MCP servers, and user-supplied model keys introduce additional processors or access paths that require separate review.[14][21][34]

On July 27, 2026, GitHub separated access to the Copilot app from the Copilot CLI access policy. On the same date, GitHub added the Copilot app and cloud agent as supported clients for enterprise-managed settings. These controls operate alongside repository permissions and other Copilot policies.[56][57]

The data terms differ by license and agreement. GitHub's terms effective April 27, 2026 state that for Copilot obtained through an individual license, GitHub may use inputs and outputs to train and develop AI models unless the user opts out. That license does not permit GitHub or its affiliates to share those inputs or outputs with third-party AI model providers for the providers' independent model training. The opt-out in this section covers AI-feature inputs and outputs; it does not alter the broader content licenses in sections D.4 through D.8 or cover repository content that was not supplied as an AI-feature input or output. Organizations purchasing through volume licensing or another customer agreement are governed by the applicable agreement and product terms rather than the individual-license rule.[23]

GitHub's March 2026 Generative AI Services Terms for covered volume customers state that GitHub will not use customer inputs or outputs to train generative AI models unless the customer provides documented instructions to do so. Retention details vary by service and are described in product documentation; the module forms part of the applicable volume agreement and data-protection terms. These contractual distinctions are more precise than a blanket claim that Copilot either always trains or never trains on customer code.[35]

Users are responsible for deciding whether an output is suitable and whether its use complies with law and licenses. Some customer agreements may provide defense obligations for eligible claims, subject to their terms and required mitigations. This should not be summarized as universal indemnity for every Copilot user or every output.[23][35]

## Adoption

Microsoft reports Copilot adoption in earnings materials, but its metrics use different populations and dates. In its fiscal 2026 second-quarter call, Microsoft said GitHub Copilot had more than 4.7 million paid subscribers, up 75 percent year over year. In the following quarter, Microsoft said nearly 140,000 organizations were using Copilot and that enterprise subscribers had nearly tripled year over year. These are company-reported measures, not independently audited estimates, and paid subscribers should not be added to organization counts.[36][37]

Earlier "users," "all-time users," "paid subscribers," "businesses," and "organizations" figures are not interchangeable. A historical total may include free or inactive accounts, while a paid-subscriber figure measures a different population. Comparisons should use the label and reporting date supplied by Microsoft or GitHub.[36][37]

## Research on productivity

Research has found productivity gains in several settings, but the magnitude depends on the task, population, outcome, and product generation. The studies below do not establish that every developer or every project becomes faster.[38][39]

| Study | Design | Reported result | Important limit |
|---|---|---|---|
| Peng et al. (2023) | Controlled experiment in which recruited developers implemented a JavaScript HTTP server | The Copilot group completed the task 55.8 percent faster | One standardized task; the study did not measure code quality and included GitHub and Microsoft-affiliated authors |
| Cui et al. (2026) | Three randomized field experiments at Microsoft, Accenture, and an anonymous Fortune 100 company, covering 4,867 developers | Precision-weighted analysis estimated a 26.08 percent increase in weekly completed pull requests among users, with a 10.3 percent standard error | Individual experiments were noisy, adoption was incomplete, and output counts are not a complete quality measure |

The Peng experiment offers a clear causal estimate for its timed task, not a universal productivity multiplier. The authors explicitly cautioned that results could vary across tasks and languages and that their experiment did not assess code quality. Extrapolating 55.8 percent to a whole engineering organization would ignore review, coordination, maintenance, and production operations.[38]

The later field experiments studied ordinary company work over periods of two to eight months. Their combined estimate is more representative of professional repositories than a short laboratory task, but the authors reported large standard errors and differences across companies. They also noted that a higher number of builds could reflect either more productive iteration or more trial and error. The study is evidence of an average effect under its deployments, not proof of an identical effect for all teams.[39]

## Security and code quality

GitHub's responsible-use documentation says Copilot can produce inaccurate, insecure, or outdated code. It recommends human review, testing, and security tooling. Academic results reinforce that advice, but the studies examined different versions and tasks and should not be merged into one failure rate.[4][40][41][42][43]

| Study | What was tested | Main finding | Scope caveat |
|---|---|---|---|
| Pearce et al. (2022) | An early Copilot preview on 89 security-oriented scenarios, producing 1,689 programs | About 40 percent of generated programs were classified as vulnerable | Artificial prompts, three languages, and a 2021 black-box product version |
| Perry et al. (2023) | 47 participants completing five security tasks with or without a custom Codex-based assistant | Assistant users produced less secure answers on several tasks and were more likely to rate insecure answers as secure | The interface used Codex-davinci-002 and was not the retail Copilot client |
| Fu et al. (2025) | 733 attributed snippets from public GitHub projects, 672 attributed to Copilot and the remainder to two other assistants | Static analysis and manual checking found weaknesses in about 30 percent of the combined sample across 43 CWE categories | Attribution came from project evidence; the sample was Python and JavaScript heavy, game projects were overrepresented, and analyzers can miss or falsely flag issues |
| Amro and Alalfi (2026) | Copilot code review on seven deliberately vulnerable benchmark datasets | Fewer than 20 comments were produced and known SQL injection, cross-site scripting, and insecure-deserialization examples were frequently missed | The tests used the code-review feature while it was in public preview in early 2025, not every later model or configuration |

The Pearce study is useful historical evidence that completion context can elicit insecure code. Its authors also warned that Copilot was a changing black-box service and that their targeted scenarios did not reproduce the full context of real software. The result should therefore be dated and scoped rather than presented as a current Copilot-wide vulnerability rate.[40]

Perry and colleagues studied the interaction between users and generated code. Their finding that some users were both less secure and more confident is relevant to review practices, but the assistant was a research interface built on an OpenAI Codex model. It did not test current Copilot agents or code review.[41]

Fu and colleagues examined snippets developers had attributed to Copilot, CodeWhisperer, or Codeium in public repositories. The study combined static analyzers with manual checks and reported a varied set of weaknesses. Its limitations included an uneven project sample, incomplete analyzer coverage, potential false positives, and uncertainty about how representative attributed snippets were of all generated code.[42]

Amro and Alalfi tested the code-review feature directly against labeled vulnerable projects and test cases. The feature often returned style or spelling comments while missing the labeled vulnerabilities. This supports treating Copilot review as an additional reviewer, not a security gate. Because the product was still in public preview at the time of testing, the study does not establish a permanent rate for later versions.[43]

## Copyright and litigation

Copilot has been disputed because its early models were trained on public source code and generated suggestions could sometimes resemble public code. Legal questions include training, reproduction, license attribution, and copyright-management information. GitHub's public-code reference feature and customer terms address parts of this risk, but neither resolves the underlying legal questions for every jurisdiction or use.[22][23]

In *Doe 1 et al. v. GitHub, Inc. et al.*, anonymous plaintiffs sued GitHub, Microsoft, and OpenAI in the United States District Court for the Northern District of California in November 2022. In January 2024, the court dismissed several state-law claims with prejudice and dismissed claims under sections 1202(b)(1) and 1202(b)(3) of the Digital Millennium Copyright Act with leave to amend, while allowing some plaintiffs' damages standing to remain.[44][45]

The district court later concluded that the amended section 1202 claims required identical output and dismissed them. On September 27, 2024, it certified the identicality question for interlocutory appeal and stayed district-court proceedings while the Ninth Circuit reviewed the issue. The order described the question as whether those provisions impose an identicality requirement and found substantial ground for disagreement.[46]

The Ninth Circuit held oral argument in appeal number 24-7700 on February 11, 2026. The court's official video archive records the hearing. No published opinion resolving the appeal appeared in the Ninth Circuit's official opinion index by the July 28, 2026 evidence cutoff, so the case should not be described as finally decided.[47][48]

## Responsible use

Copilot output should be reviewed as untrusted proposed code. A developer should understand the change, run relevant tests, check failure paths and boundary conditions, review dependencies and licenses, and apply the same security controls required for human-written code. High-risk changes may require threat modeling, static and dynamic analysis, dependency and secret scanning, and review by a person accountable for the affected system.[4]

Agentic workflows add operational risk because they can modify multiple files and invoke tools or commands. Repositories should use least-privilege credentials, protected branches, required reviews, limited Actions permissions, reproducible environments, and budgets. Session logs and diffs should be inspected before merge. An agent's successful test run does not prove that the test suite is complete.[5][6]

Sensitive repositories need a surface-by-surface data review. Content exclusions have documented gaps, user-supplied model keys can move processing to another provider, MCP servers can expose external systems, and cloud-agent sessions use GitHub-hosted environments. Contract terms, organization policy, and technical controls should be evaluated together.[7][9][14][21][23][35]

## Alternatives

AI coding assistants differ in where they operate and who provides the service. [Cursor](https://aiwiki.ai/wiki/cursor) is an editor-centered product, [Claude Code](https://aiwiki.ai/wiki/claude_code) is primarily a terminal agent, [Amazon Q](https://aiwiki.ai/wiki/amazon_q) emphasizes Amazon Web Services development, [Tabnine](https://aiwiki.ai/wiki/tabnine) offers enterprise deployment choices, and [Gemini Code Assist](https://aiwiki.ai/wiki/gemini_code_assist) integrates with Google's developer and cloud tooling. Feature, model, price, and data-policy comparisons change frequently and should be based on each vendor's current documentation rather than a fixed ranking.[49][50][51][52][53]

Copilot combines editor clients with GitHub repositories, pull requests, Actions, and organization policy. Repository permissions, Actions settings, Copilot policy, AI Credit budgets, and external-tool permissions jointly determine the effective boundary.[5][8][34]

## See also

- [OpenAI Codex](https://aiwiki.ai/wiki/openai_codex)
- [Large Language Model](https://aiwiki.ai/wiki/large_language_model)
- [Software Development](https://aiwiki.ai/wiki/software_development)
- [Model Context Protocol](https://aiwiki.ai/wiki/model_context_protocol)
- [Copyright and AI](https://aiwiki.ai/wiki/copyright)
- [Cursor (code editor)](https://aiwiki.ai/wiki/cursor)
- [Claude Code](https://aiwiki.ai/wiki/claude_code)

## References

1. Nat Friedman. "Introducing GitHub Copilot: your AI pair programmer." The GitHub Blog, June 29, 2021. https://github.blog/news-insights/product-news/introducing-github-copilot-ai-pair-programmer/
2. GitHub. "GitHub Copilot is generally available to all developers." The GitHub Blog, June 21, 2022. https://github.blog/news-insights/product-news/github-copilot-is-generally-available-to-all-developers/
3. GitHub Docs. "GitHub Copilot features." https://docs.github.com/en/copilot/get-started/features
4. GitHub Docs. "Responsible use of GitHub Copilot Chat in GitHub." https://docs.github.com/en/copilot/responsible-use/chat-in-github
5. GitHub Docs. "About Copilot cloud agent." https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent
6. GitHub Docs. "About GitHub Copilot CLI." https://docs.github.com/en/copilot/concepts/agents/copilot-cli/about-copilot-cli
7. GitHub Docs. "GitHub Copilot app." https://docs.github.com/en/copilot/concepts/agents/github-copilot-app
8. GitHub Docs. "About GitHub Copilot code review." https://docs.github.com/en/copilot/concepts/agents/code-review
9. GitHub Docs. "Content exclusion." https://docs.github.com/en/copilot/concepts/context/content-exclusion
10. GitHub Docs. "About GitHub Copilot code suggestions." https://docs.github.com/en/copilot/concepts/completions/code-suggestions
11. GitHub. "Next edit suggestions, agent mode, and prompt files for GitHub Copilot in VS Code." GitHub Changelog, February 6, 2025. https://github.blog/changelog/2025-02-06-next-edit-suggestions-agent-mode-and-prompts-files-for-github-copilot-in-vs-code-january-release-v0-24/
12. GitHub. "GitHub Copilot is moving to usage-based billing." The GitHub Blog, April 27, 2026. https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/
13. GitHub. "GitHub Copilot Chat now generally available for organizations and individuals." The GitHub Blog, December 29, 2023. https://github.blog/news-insights/product-news/github-copilot-chat-now-generally-available-for-organizations-and-individuals/
14. GitHub Docs. "About Model Context Protocol." https://docs.github.com/en/copilot/concepts/context/mcp
15. GitHub. "GitHub Copilot: Meet the new coding agent." The GitHub Blog, May 19, 2025. https://github.blog/news-insights/product-news/github-copilot-meet-the-new-coding-agent/
16. GitHub. "Copilot coding agent is now generally available." GitHub Changelog, September 25, 2025. https://github.blog/changelog/2025-09-25-copilot-coding-agent-is-now-generally-available/
17. GitHub. "Research, plan, and code with Copilot cloud agent." GitHub Changelog, April 1, 2026. https://github.blog/changelog/2026-04-01-research-plan-and-code-with-copilot-cloud-agent/
18. GitHub. "GitHub Copilot CLI is now generally available." GitHub Changelog, February 25, 2026. https://github.blog/changelog/2026-02-25-github-copilot-cli-is-now-generally-available/
19. GitHub. "GitHub Copilot app available to all." GitHub Changelog, July 7, 2026. https://github.blog/changelog/2026-07-07-github-copilot-app-available-to-all/
20. GitHub Docs. "Repository indexing for GitHub Copilot." https://docs.github.com/en/copilot/concepts/context/repository-indexing
21. GitHub Docs. "Excluding content from GitHub Copilot." https://docs.github.com/en/copilot/how-tos/configure-content-exclusion/exclude-content-from-copilot
22. GitHub Docs. "Finding public code that matches GitHub Copilot suggestions." https://docs.github.com/en/copilot/concepts/completions/code-referencing?tool=webui
23. GitHub. "GitHub Terms of Service." Effective April 27, 2026. https://docs.github.com/en/site-policy/github-terms/github-terms-of-service
24. Mark Chen et al. "Evaluating Large Language Models Trained on Code." arXiv, 2021. arXiv:2107.03374. https://arxiv.org/abs/2107.03374
25. GitHub. "GitHub Copilot for Business is now available." The GitHub Blog, February 14, 2023. https://github.blog/news-insights/product-news/github-copilot-for-business-is-now-available/
26. GitHub. "GitHub Copilot X: The AI-powered developer experience." The GitHub Blog, March 22, 2023. https://github.blog/news-insights/product-news/github-copilot-x-the-ai-powered-developer-experience/
27. GitHub. "GitHub Copilot Enterprise is now generally available." The GitHub Blog, February 27, 2024. https://github.blog/news-insights/product-news/github-copilot-enterprise-is-now-generally-available/
28. GitHub. "Universe 2024: Copilot advances." The GitHub Blog, October 29, 2024. https://github.blog/news-insights/product-news/universe-2024-previews-releases/
29. GitHub. "Announcing GitHub Copilot Free." GitHub Changelog, December 18, 2024. https://github.blog/changelog/2024-12-18-announcing-github-copilot-free/
30. GitHub Next. "GitHub Copilot Workspace." Technical-preview archive and sunset notice. https://githubnext.com/projects/copilot-workspace/
31. GitHub Docs. "GitHub Copilot feature matrix." https://docs.github.com/en/copilot/reference/copilot-feature-matrix
32. GitHub Docs. "Supported AI models in GitHub Copilot." https://docs.github.com/en/copilot/reference/ai-models/supported-models?library=true
33. GitHub Docs. "Plans for GitHub Copilot." https://docs.github.com/en/copilot/get-started/plans
34. GitHub Docs. "GitHub Copilot billing." https://docs.github.com/en/billing/concepts/product-billing/github-copilot-billing
35. GitHub. "GitHub Generative AI Services Terms." March 2026. https://github.com/customer-terms/github-generative-ai-services-terms
36. Microsoft. "FY 2026 Q2 earnings." Investor Relations, January 28, 2026. https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q2
37. Microsoft. "FY 2026 Q3 earnings." Investor Relations, April 29, 2026. https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q3
38. Sida Peng, Eirini Kalliamvakou, Peter Cihon, and Mert Demirer. "The Impact of AI on Developer Productivity: Evidence from GitHub Copilot." arXiv, 2023. arXiv:2302.06590. https://arxiv.org/abs/2302.06590
39. Kevin Zheyuan Cui, Mert Demirer, Sonia Jaffe, Leon Musolff, Sida Peng, and Tobias Salz. "The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers." Management Science, published online February 27, 2026. https://doi.org/10.1287/mnsc.2025.00535
40. Hammond Pearce, Baleegh Ahmad, Benjamin Tan, Brendan Dolan-Gavitt, and Ramesh Karri. "Asleep at the Keyboard? Assessing the Security of GitHub Copilot's Code Contributions." 2022 IEEE Symposium on Security and Privacy, 2022. https://doi.org/10.1109/SP46214.2022.9833571
41. Neil Perry, Megha Srivastava, Deepak Kumar, and Dan Boneh. "Do Users Write More Insecure Code with AI Assistants?" Proceedings of the 2023 ACM SIGSAC Conference on Computer and Communications Security, 2023. https://doi.org/10.1145/3576915.3623157
42. Yujia Fu, Peng Liang, Amjed Tahir, Zengyang Li, Mojtaba Shahin, Jiaxin Yu, and Jinfu Chen. "Security Weaknesses of Copilot-Generated Code in GitHub Projects: An Empirical Study." ACM Transactions on Software Engineering and Methodology, 2025. https://arxiv.org/abs/2310.02059
43. Amena Amro and Manar H. Alalfi. "GitHub's Copilot Code Review: Can AI Spot Security Flaws Before You Commit?" Proceedings of Machine Learning Research, volume 318, 2026, pages 825-832. https://proceedings.mlr.press/v318/amro26a.html
44. United States District Court for the Northern District of California. "Doe 1 et al. v. GitHub, Inc. et al., Case No. 4:22-cv-06823-JST." Official case page. https://cand.uscourts.gov/cases-e-filing/cases/422-cv-06823-jst/doe-1-et-al-v-github-inc-et-al
45. United States District Court for the Northern District of California. "Order granting in part and denying in part motion to dismiss," Document 195, filed January 22, 2024. https://www.govinfo.gov/content/pkg/USCOURTS-cand-4_22-cv-06823/pdf/USCOURTS-cand-4_22-cv-06823-1.pdf
46. United States District Court for the Northern District of California. "Order granting motion to certify order for interlocutory appeal and motion to stay pending appeal," Document 282, September 27, 2024. https://www.govinfo.gov/content/pkg/USCOURTS-cand-4_22-cv-06823/pdf/USCOURTS-cand-4_22-cv-06823-3.pdf
47. United States Court of Appeals for the Ninth Circuit. "Case 24-7700 oral argument video archive," February 11, 2026. https://www.ca9.uscourts.gov/cases/streams-videos/archive?case=24-7700&hearingDate=2026-02-11
48. United States Court of Appeals for the Ninth Circuit. "Opinions search." Official opinions index queried for case 24-7700 through the evidence cutoff. https://rf9ecr1p0i.execute-api.us-west-2.amazonaws.com/prod/opinions/search?q=24-7700
49. Anysphere. "Cursor features." https://www.cursor.com/features
50. Anthropic. "Claude Code overview." https://docs.anthropic.com/en/docs/claude-code/overview
51. Amazon Web Services. "Amazon Q Developer." https://aws.amazon.com/q/developer/
52. Tabnine. "Tabnine AI code assistant." https://www.tabnine.com/
53. Google Cloud. "Gemini Code Assist overview." https://cloud.google.com/products/gemini/code-assist
54. GitHub. "GitHub Copilot is now available for invoiced GitHub Enterprise customers." GitHub Changelog, December 7, 2022. https://github.blog/changelog/2022-12-07-github-copilot-is-now-available-for-invoiced-github-enterprise-customers/
55. GitHub. "GitHub Copilot general availability in the CLI." GitHub Changelog, March 21, 2024. https://github.blog/changelog/2024-03-21-github-copilot-general-availability-in-the-cli/
56. GitHub. "Manage GitHub Copilot app access with a dedicated policy." GitHub Changelog, July 27, 2026. https://github.blog/changelog/2026-07-27-manage-github-copilot-app-access-with-a-dedicated-policy/
57. GitHub. "Enterprise managed settings in the GitHub Copilot app and Copilot cloud agent." GitHub Changelog, July 27, 2026. https://github.blog/changelog/2026-07-27-enterprise-managed-settings-now-apply-to-the-github-copilot-app/