TALO

TALO audits fraud, KYC and AML models for banks and fintechs without sending customer data anywhere. You upload a model’s decision log. Deterministic statistics (PSI, KS test, per-channel alert-rate tests, label delay) find drift and turn it into numbered evidence, with a per-channel view of where drift is and isn’t. Meta’s Muse Spark then writes the audit report. Every claim cites evidence, code rejects any number not in that evidence, and a second agent attacks the rest. Weak claims are rewritten once or marked “insufficient evidence”. Possible causes are kept separate as unverified hypotheses with a next check. A red-team button injects fabricated claims, and an Egress Ledger shows every request sent to the model, with zero raw customer rows. It includes accounts, a column-mapping wizard, a printable fingerprinted report, a live demo and an evaluation harness.