Enoki tests AI agents for security and safety weaknesses. It attacks your agent the way an adversary would, adapts to how the agent responds, and reports each weakness it finds with the exact attack and the agent's response.
With Enoki in ChatGPT you can:
- Register an agent endpoint as a target and keep its profile current.
- Start an assessment, on everything or on the categories a change touched, and follow it to the end.
- Read findings with the full attack transcript, so you can fix the weakness in your code or configuration.
- Close findings, and re-run a finding's attacks against your live agent to confirm a fix worked.
A typical loop: you change your agent's prompt or tools, ask ChatGPT to test the change, read what Enoki found, fix it, and confirm the fix.
You sign in to Enoki and approve the connection, choosing what ChatGPT may do: read only, change targets and findings, or also start assessments. You can change or revoke that at any time under Settings > Connected apps.
Enoki needs an Enoki workspace; an owner or admin of the workspace approves the connection. Register a test or staging deployment of your agent rather than production, because an assessment sends adversarial messages to it.
How to use
- 1Add it in ChatGPT
Open Enoki in ChatGPT's Plugin directory (button above), select + to add it and sign in if asked.
- 2Use it in a chat
Type @ in the composer and pick Enoki, or start your message with its name. ChatGPT may also suggest it when it fits what you ask.
Skills 1
- test-agent-change
Test an AI agent with Enoki after a change to it, and re-verify fixes. Use when the user has changed an agent's system prompt, tools, model, data sources or guardrails and wants to know what the change broke, asks to test or red-team their agent, or says a finding is fixed and wants that confirmed.
Things to ask
- “Test my support agent with Enoki.”
- “What did the last Enoki run find on my agent?”
- “I fixed the finding Enoki reported. Check that it is fixed.”