Q2 says its embedded banking assistant has received AIUC-1 certification after a third-party controls audit and adversarial testing. The November 5 Q2 announcement describes tests of data isolation, permissions, interaction logging and compliance controls.

Q2 Assistant is designed to place specialized AI agents inside the provider's banking products. The company says those agents operate within roles and permissions established by each financial institution. The certification announcement names six assessment domains: data and privacy, security, safety, reliability, accountability and societal impact.

Certification is evidence, not the whole decision

The AIUC-1 program describes certification as evidence that an AI system was developed and deployed against its current security, safety and reliability practices. Q2 says its review included prompt injection, jailbreaks, unauthorized actions, data leakage, hallucination and unsafe tool calls, with quarterly adversarial retesting planned.

The announcement does not publish the underlying audit report, test cases, exception list, sampled configurations or a certificate scope that a credit union can compare with its exact Q2 products. The AIUC public updates page did not provide a separate Q2 announcement when checked November 7. The evidence available publicly is therefore the provider's certification statement plus general information about the standard, not an independently inspectable test record.

Ask what is inside the assessed boundary

A credit union can use the certification to make its vendor review more specific. Which Q2 agents and versions were in scope? Which tools can each agent call? Are institution-specific extensions covered? What happens when an agent requests an action outside its permission boundary? How quickly do new models, prompts or tool connections trigger retesting?

The institution also needs its own evidence. Configuration choices, staff access, member data, connected systems and human approvals sit outside a generic vendor claim. Before activation, teams should map each permitted action to a named owner, log source and reversal or shutdown path. After activation, they should test whether local permissions and logs behave as documented.

Our coverage of the FSB's AI lifecycle practices places vendor evidence inside a broader governance process. The bank AI control-gap survey shows why a policy document alone is not proof that operational controls work.

The useful change is the availability of another evidence artifact for a risk committee. The limit is equally important: certification provenance does not establish that every Q2 deployment, institution configuration or future agent behavior is safe.

Follow consequential credit-union developments. Explore source-backed operating coverage and weekly web briefings.

Browse web briefings