# AI Security for AppSec Teams Checklist

Use this checklist before scaling secure coding assistants, AI code review, dependency triage, GitHub security automation, or AI-assisted CI/CD changes.

## 1. Coding Assistant Scope

- [ ] Define which repositories AI coding assistants may read, edit, test, and summarize.
- [ ] Block assistants from reading secrets, production credentials, customer data, and sensitive incident artifacts.
- [ ] Require explicit approval before AI agents open pull requests, change workflows, publish packages, or deploy code.
- [ ] Keep AI-generated changes in branches with normal review and merge controls.

## 2. GitHub and Repository Guardrails

- [ ] Require branch protection, code owners, required reviews, code scanning, secret scanning, and dependency review.
- [ ] Pin sensitive GitHub Actions to full commit SHAs.
- [ ] Set default workflow token permissions to read-only where possible.
- [ ] Avoid exposing secrets to forked pull requests or untrusted workflows.
- [ ] Prefer OIDC federation over static cloud credentials.

## 3. Evidence-Based AI Review

- [ ] Require AI findings to cite files, lines, functions, tests, package metadata, advisories, and exploitability reasoning.
- [ ] Treat hallucinated vulnerabilities and hallucinated fixes as expected failure modes.
- [ ] Use AI to draft tests and patches, but require human review for security impact.
- [ ] Confirm remediation with tests, scanners, and code owners before merge.

## 4. Dependency and Supply Chain Risk

- [ ] Use AI to cluster dependency findings by reachability, exploitability, and package ownership.
- [ ] Keep package install scripts away from production credentials.
- [ ] Use disposable dev containers for unfamiliar repositories or supply-chain investigations.
- [ ] Rotate tokens after suspected malicious package execution.

## 5. CI/CD Safety

- [ ] Review any AI-authored workflow change as security-sensitive.
- [ ] Separate test workflows from release workflows.
- [ ] Require environment protection for production deployment.
- [ ] Log assistant, prompt, branch, PR, reviewer, and deployment metadata.

## 6. Metrics

- [ ] Track vulnerability remediation time, reopened fixes, false-positive reduction, review latency, and AI-authored change rejection rate.
- [ ] Review whether AI is improving secure delivery or only increasing change volume.

Related HackWednesday pages:

- https://hackwednesday.com/ai-security-for-appsec
- https://hackwednesday.com/best-ai-coding-assistants-for-security-teams
- https://hackwednesday.com/resources/github-ai-security-skills-for-appsec-teams
- https://hackwednesday.com/blog/github-security-best-practices-for-teams