GitHub lets teams assign Dependabot alerts to AI agents for remediation

GitHub on April 7, 2026 announced that Dependabot alerts can now be assigned directly to AI coding agents for remediation. The feature targets cases where a vulnerability cannot be fixed with a simple package bump and instead needs code changes across the repository. GitHub says teams can route an alert to Copilot, Claude, or Codex from the alert detail page and receive a draft pull request with a proposed fix.

According to GitHub, the assigned agent analyzes the security advisory and the way the vulnerable dependency is actually used in the repository. It then opens a draft pull request and attempts to resolve test failures introduced by the dependency update. Multiple agents can be assigned to the same alert, which lets maintainers compare remediation strategies rather than relying on a single model’s proposal.

• Major version upgrades with breaking API or type changes.
• Package downgrades when a dependency is compromised and no patched release is available.
• More complex pull requests that fall outside Dependabot’s existing rule-based update engine.

The announcement is notable because it connects two parts of the software supply-chain workflow that are often separate today. Dependabot already handles the structured detection and routine update path. Coding agents are being inserted where security work becomes repository-specific and expensive in engineering time. GitHub is effectively turning the alert itself into the handoff point between vulnerability detection and code remediation.

GitHub also made the limitation explicit: AI-generated fixes are not guaranteed to be correct. The company says teams should always review the proposed pull request, confirm that tests pass, and verify that the security issue is actually resolved before merging. Access requires GitHub Code Security and a Copilot plan with coding agent access. For security teams, the feature is important less because it removes humans from the loop and more because it could reduce the backlog of alerts that stall when remediation needs real code surgery.