AI-Assisted Secure Coding (Appendix C) · Adversarial AI Detection in Inbound Contributions
Contain adversarial contributions automatically
Verify that confirmed or high-confidence adversarial contributions trigger automated containment: block the PR, quarantine the fork, suspend the contributor, notify maintainers, and freeze affected workflow files. Triage decisions feed back into detection tuning.
OWASP AISVS v1.01, requirement AC.13.5
In plain English
When a contribution is confirmed or very likely to be an attack, contain it automatically. Block the PR, quarantine the fork, suspend the contributor, tell the maintainers and freeze the affected workflow files. Then feed what you learned back into detection. This is a Level 3 control.
What good looks like
- High-confidence detections trigger containment without waiting for a person.
- Containment covers all five actions in the requirement, as far as your platform allows.
- Maintainers are notified at once and can reverse a wrong call.
- Triage outcomes are used to tune detection.
How teams usually do it
- Automation driven by high-confidence alerts, using the code platform's API to close and lock the PR, block the user from the organisation and tighten rulesets on workflow paths.
- A lower-confidence tier that holds the PR and asks a maintainer.
- A post-incident step that records whether the detection was right.
Suggested evidence
- Automation configuration.
- A test or real containment record with timings.
- Evidence of detection tuning from triage.
Common mistakes
- Waiting for a maintainer to wake up before blocking.
- Automatic suspension on weak signals, which drives away real contributors.
- Containing the PR and leaving its already-triggered workflow runs going.
Not sure which requirements apply to your system? Take the assessment, or browse every guide in this chapter.
Requirement text from the OWASP AI Security Verification Standard, licensed under CC BY-SA 4.0. This guidance is shared under the same licence.