Daniel Neuhaus

A Sign Is Not a Lock: Why AI Agents Need Access Limits, Not Just Instructions

An AI agent that drafts a customer review may also be able to change the case status. An agent's promise to wait for approval is evidence about model behaviour. It does not show that the action is prevented. This article proposes a two-pass test. The first pass exercises the perm...

Right Answer, Wrong Reason: Testing Source References and Responses to Evidence Changes

An AI system can extract the right value from the wrong place in a document. When a bank relies on that value, checking the answer alone can miss the source error. This article proposes a separate source check followed by controlled changes to relevant and irrelevant evidence. Th...

Cheap Checks, Costly Decisions: What AI Really Costs in Anti-Money-Laundering Reviews

Banks use AI to prepare anti-money-laundering reviews of their customers. Even when the automated check becomes cheaper, analysts may have to spend more time verifying the evidence and recording who was authorised to take the decision. This article proposes a comparison protocol....