GitHub: Automatic code security validation is expanding to third-party agents

On June 9, 2026, GitHub announced the extension of automatic code security validation to all third-party coding agents working with the platform. Protection that was previously reserved exclusively for the GitHub Copilot cloud agent now applies to external agents as well — including Anthropic Claude and OpenAI Codex — that open pull requests in user repositories.

Context: Copilot cloud agent received protection in October 2025

GitHub first introduced automatic security validation of generated code in October 2025, at the time only for its own Copilot cloud agent. In the first eight months of deployment, that protection according to GitHub proactively prevented hundreds of potential security issues and vulnerabilities in user repositories.

Simultaneously, the ecosystem of third-party agents integrating via the GitHub API grew strongly. Development teams increasingly use external agents that autonomously write code, commit changes, and open pull requests. This dynamic created a clear gap: Copilot’s code was being scanned, while code from external agents — flowing through the same repositories and the same workflows — was not.

Which security checks does the agent automatically perform on generated code?

Every pull request that a third-party agent’s generated code submits to a GitHub repository passes through three automated security steps:

CodeQL vulnerability scanning GitHub’s static analysis tool reviews the generated code looking for patterns indicative of known security vulnerabilities — from SQL injection and cross-site scripting to unsafe file operations, depending on the programming language. CodeQL understands code semantics rather than just text patterns, giving it a lower false-positive rate than simple regex checks.

Dependency check against the Advisory database Every new or modified dependency is automatically compared against the GitHub Advisory Database — a database of known vulnerabilities in open-source packages (CVE records). If an agent introduces a package with an active vulnerability, the check flags it immediately and blocks finalization of the pull request until the issue is resolved.

Exposed secret scanning The scan passes through the generated code looking for API keys, OAuth tokens, passwords, and other sensitive data (secrets) that have accidentally or mistakenly ended up in the code. This measure is especially relevant for agents that automatically generate configuration files or code examples — a common source of unintentional data leaks.

Agent response to found issues

Validation is not passive reporting — it is integrated into the agentic workflow. When a vulnerability, problematic dependency, or exposed secret is discovered, the agent attempts to automatically fix the problem before finalizing the pull request. This may include replacing vulnerable packages with safer versions, removing exposed keys, or adjusting code according to CodeQL recommendations.

Only if automatic remediation is not possible is the issue escalated to the development team, which must intervene manually. This approach means that a significant portion of security corrections remain within the agentic workflow — vulnerabilities do not reach the code review stage or later development phases where they are more expensive to fix.

No special licenses required

The extension is enabled by default and follows the Copilot settings the administrator has already configured for the repository. No manual activation, separate configuration for each agent, or subscription upgrade is needed.

Crucially, the functional protection does not require a GitHub Advanced Security license. It is available to all users at no additional cost, which is particularly important for smaller teams and open-source projects that do not have access to advanced security tools under the Enterprise plan.

Organizations that want to fine-tune which checks are active — or exclude a specific agent from validation — can do so through the settings described in the GitHub documentation for configuring agent settings.

---

By extending security validation to all agents, GitHub is building a consistent security layer that does not depend on which AI model or agent generates the code. For development teams that have started adopting external agents in their CI/CD workflow, this upgrade means the same protection standards now cover the entire workflow — automatically, without additional cost or manual setup.