Skip to main content

Threat Modeling AI-Assisted Code

Lightweight Model (10 Minutes)​

  1. Identify trust boundaries touched by the change.
  2. List attacker entry points (API, UI, async jobs, third-party callbacks).
  3. Map failure impact (data leak, privilege escalation, service disruption).
  4. Define one preventive control and one detective control per high-risk path.
  5. Add verification tests before merge.

AI-Specific Threat Patterns​

  • Hallucinated library calls with insecure defaults
  • Missing input constraints from generated handlers
  • Broken permission checks due to copied patterns
  • Hidden data exposure in debug logging

Deliverable​

Add a short threat note in the PR:

  • Threat surface changed
  • Top risks
  • Controls added
  • Residual risks accepted