THE THREAT MODEL

The attacks that trick an agent into moving money it should not.

An AI agent that can pay is an AI agent that can be tricked into paying. These are the concrete attacks, each mapped to the OWASP Agentic Top 10 class it belongs to. Fidacy stops every one of them deny-by-default, and issues Bitcoin-anchored proof, third-party verifiable, of exactly what was and was not authorized.

targeted every day · deny-by-default
3,087 Actions gated, all time
3,441 Signed verdicts issued

Live from the engine, re-fetch them yourself at api.fidacy.com/v1/pulse. Early traffic includes evaluation and testing, we do not dress it up as real-world attacks.

{{ t.id }} {{ t.owasp }}
{{ t.count }}

{{ t.name }}

THE ATTACK

{{ t.attack }}

THE BLOCK

{{ t.block }}
Run these attacks against the live engine → Then try to forge the proof →

Fidacy does not promise your agent will never be attacked. It promises the unauthorized payment does not settle, and that you can prove, to a counterparty, an auditor or an insurer, exactly what was authorized. Deny-by-default means the failure mode is money that does not move, never money that moves wrong.