AI-Assisted Secure Coding (Appendix C) · AI-Assisted Secure-Coding Workflow
Name the AI threats you defend against
Verify that the workflow names the adversarial-AI threat scenarios it is built to mitigate. The list should cover prompt injection delivered through PR content, AI-generated supply-chain payloads, autonomous agents approving their own work, fork-PR secret exfiltration, and compromise of the model supply chain.
OWASP AISVS v1.01, requirement AC.1.3
In plain English
List the attacks on AI-assisted development that your workflow is built to stop. The standard names five: prompt injection through pull request content, AI-generated supply-chain payloads, agents approving their own work, secret theft through fork pull requests, and compromise of the model supply chain. A named threat can be matched to a control. An unnamed one gets missed.
What good looks like
- The workflow names its threat scenarios, including the five in the requirement.
- Each threat maps to at least one control, with an owner.
- The list is reviewed when tools or pipelines change.
- Public incidents involving AI coding tools are checked against the list.
How teams usually do it
- A short threat model document for the AI-assisted development workflow, kept beside the workflow itself.
- A table with threat, entry point, control and evidence as its columns.
- A standing item in security reviews to add new attack reports.
Suggested evidence
- The threat list with mapped controls.
- The record of the last review and what changed.
Common mistakes
- A list of threats with no link to the controls that address them.
- Copying the five examples and never considering what is particular to your own pipeline.
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.