# CAIN-A75-050 Agent Interaction Risk Review

**Verdict:** VERIFIED

> Recomputes agent-interaction decisions: allowlisted hashed signals with no raw behaviour, recomputed risk and agent scores, and ALLOW only with low risk, no fraud hold and a valid mandate for an agent. It does not grade the model.

| Session | Verdict | Interactions | Failed checks |
|---|---|---|---|
| 0 bank:demo | SOUND | 2 |  |

## What this proves and what it does not
- Proves: every signal is on the allowlist, committed by SHA-256 and stored without raw behaviour; the stated risk score and agent flag equal the recomputed weighted sums; every ALLOW had risk at or under the threshold, no fraud hold, and (for an agent) a valid human-issued mandate for this account, payee, amount and time; and the package's claims match.
- Does not prove: that the signal model detects agents or fraud well, or that the hashed raw sources exist. It is a decision-record review, not a fraud model.
- Verify offline: `python3 verify_offline.py run.json --pinned-key 1NH6fvm0WPM6D3RYrpcQeRhzzG+qMib4SEWeRwDJ41Y=`
