AI-Assisted Secure Coding (Appendix C) · Compromise Containment & Automated Remediation
Exercise automated remediation yearly
Verify that automated remediation is tested in tabletop or live-fire exercises at least once a year. The scenarios include a prompt-injected reviewer bot, fork-PR secret exfiltration, and an AI-generated malicious workflow file.
OWASP AISVS v1.01, requirement AC.14.5
In plain English
Test your automated remediation at least once a year, in a tabletop or a live exercise. Include three scenarios: a review bot taken over by prompt injection, secrets stolen through a fork PR, and a malicious workflow file generated by AI. This is a Level 3 control.
What good looks like
- Exercises run at least once a year.
- They cover the three named scenarios.
- They test the automated remediation itself: revocation, rotation and containment.
- Lessons are tracked to completion.
How teams usually do it
- Tabletop sessions with engineering, security and whoever handles external notification.
- Controlled live-fire tests in a sandbox organisation, with real automation and canary secrets.
- Measuring times: detection to containment, and containment to full rotation.
Suggested evidence
- Exercise reports with timings.
- Follow-up actions and their status.
Common mistakes
- Exercises that test detection and stop before remediation.
- Tabletops where everyone agrees the automation would work and nobody runs it.
- The same scenario every year.
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.