UNAUTHORIZED ACTION TEST · NO ACCOUNT · LIVE ENGINE

Your AI agent was hijacked. Can it still act?

Assume the model has already been manipulated. Choose a controlled attack scenario. Fidacy sends the attempted action to the live authority engine, returns a signed verdict, and your browser verifies it against the public key. No account. No trust in us required.

1 · CHOOSE THE ATTACK SCENARIO
{{ s.name }} {{ s.note }}
{{ s.amount }} run →
POST /v1/assess

choose a scenario to challenge the live boundary▋

→ mandate received · {{ runningName }}
→ KYA · trust score · policy…
→ engine signing▋
{{ verdictLabel }} score {{ score }} · {{ latency }}ms
{{ reason }}
signature valid : true ✓ ← verified in your browser
EdDSA · checked against the public JWKS, not against us

This controlled simulation does not diagnose malware on your device. It proves how the live boundary responds when an AI agent presents the selected unauthorized action. For payments, a Fidacy-connected executor receives a short-lived grant only after ALLOW. A denied action receives no grant, so no money moves. Every decision enters a hash-chained audit whose head is anchored to Bitcoin. One install ships the boundary: npx -y @fidacy/mcp.

Get your free API key → Install the MCP server → Read the UCP integration →