Regula and Bernini.ai Add Identity Checks to AI Conversations

Regula and Bernini.ai announced a partnership on September 30 to bring document and biometric identity verification into customer-facing AI conversations. According to Regula's announcement, Bernini.ai's DAFNE platform can request a verification check when a customer needs to establish or reconfirm their identity.

The security significance is the point at which a conversation becomes an action. An assistant can discuss a service with a visitor without necessarily having enough evidence to disclose account information or carry out a sensitive request. The partnership puts an identity step into that transition.

Verification becomes part of the workflow

The companies describe uses in onboarding, existing-account service and higher-risk transactions. DAFNE can invoke Regula Document Reader SDK and Face SDK, then use the result alongside the business's rules to decide how the interaction continues. Regula says the identity layer includes document checks, face matching and liveness checks.

The announcement also describes specialised agents coordinated by a shared control layer. It does not provide an independently verified fraud reduction figure, deployment benchmark or error rate for the combined system. GlobalTechRanks has not tested the integration.

Knowing a person is only one part of authorisation

Our analysis is that a positive identity result should answer a limited question. It can contribute evidence about who is present, while the business still needs to establish what that person is allowed to request and whether the requested action is appropriate.

For example, a hypothetical customer might pass an identity check while asking to change an account belonging to a different organisation. The workflow would still need to establish the relationship and the relevant authority. Similarly, an authenticated person could be acting under coercion or following a scammer's instructions. Identity assurance is useful evidence, but it should not become a blanket permission to execute every subsequent request.

That creates an important design question for agent developers: which actions can proceed automatically, which require stronger evidence and which should reach a human reviewer? The answer should depend on the action and its consequences, rather than the fluency of the conversation.

What prospective buyers should inspect

A meaningful evaluation would examine failed and ambiguous cases as well as successful onboarding. Teams could ask how the workflow handles a poor-quality document, a mismatch, an interrupted check or a customer who needs an accessible alternative. They should also inspect what information is retained and which staff or systems can access it.

Those are suggested evaluation questions, not findings about shortcomings in this product. They help separate an integration demonstration from a dependable customer process.

The partnership is a concrete example of security moving inside an agent workflow. Its next test is whether organisations can use the returned identity evidence to make decisions that remain understandable, proportionate and auditable when a conversation reaches a consequential action.

Image: Regula