The problem
Companies are handing AI agents real work faster than they can prove those agents are safe. There’s no number for “how much can the agent do on its own?”, and normal audit logs can be edited by whoever owns them.
What it does
- A safety dial with a guarantee. You choose a risk budget α, and conformal risk control computes the cutoff for which actions run automatically and which go to a human. Of the actions the agent takes alone, the error rate stays at or below α, with confidence 1 − δ. If the world shifts, Veridian notices and dials the agent back.
- Tamper-proof receipts. Every decision becomes a signed, Merkle-chained record that anyone can verify offline, in the browser, without trusting the operator.
Go-to-market thinking
The beachhead is KYC periodic review and enhanced due diligence, starting with crypto exchanges and VASPs. The demo runs end to end on synthetic KYC data and is covered by 105 automated tests.