System area

AI Security

Secure development, adversarial thinking, and least-privilege authority for AI-enabled systems.

Scope

Where it helps

Use this area when models or agents can reach sensitive data, tools, users, or external systems.

Inquiry

Questions LucyAI asks

  • What can untrusted content influence?
  • Which permissions are truly required?
  • How are abuse, failure, and recovery observed?

Outputs

Artifact ledger

  1. Threat and trust-boundary model
  2. Tool-authority matrix
  3. Security validation plan

Boundaries

Control ledger

  1. Input and output validation
  2. Guarded external effects
  3. Incident and rollback paths

Evidence

Source basis

Move with intent

Choose a next step

Start with the pressure you need to resolve, or prepare a local brief that keeps the problem, constraints, and evidence together.