AIUC-1

AIUC-1 is a security, safety and reliability standard written specifically for AI agents, published by the Artificial Intelligence Underwriting Company. Certification combines an audit by an accredited firm with adversarial testing of the agent itself (prompt injection, data leakage, hallucinations, unsafe tool calls), repeated every quarter and renewed every year.

Where we are

Vaulty is not AIUC-1 certified and has not started the certification. Several design choices the standard looks for are part of Vaulty. The agent reads your systems through read-only connections. Any action that changes something, such as sending an email or writing to a system, is a separate step that a person approves. Each company's data and credentials are kept apart. Figures come from queries the user can see, and the agent's activity is logged.

What is planned

A written AI risk taxonomy; failure plans for security incidents, harmful outputs and wrong figures; automated tests for prompt injection and for isolation between companies; clear disclosure whenever content is written by AI; and documented due diligence on our model providers. We will decide on full AIUC-1 certification once these are in place, and this card will change only when an accredited auditor has signed off.