AIUC Raises $40M to Certify AI Agents Against Rogue Behavior
AIUC, a startup founded by an early Anthropic employee and a former METR COO, has raised $40 million in Series A funding. The company has developed a certification standard called AIUC-1, which tests AI agents for jailbreaks, hallucinations, and data leaks. Customers include Cursor, Lovable, Harvey, and ElevenLabs.
AIUC's prior $15 million seed round included backing from Nat Friedman's NFDG, Emergence, Terrain, and Anthropic co-founder Ben Mann. The AIUC-1 standard was developed with input from a consortium of roughly 250 enterprise security and risk leaders, and is modeled on the established SOC 2 cybersecurity framework.
The testing suite runs about 5,000 scenarios covering jailbreaks, hallucinations, and data leaks, yielding a roughly 100-page report. AI agents execute the tests and analyze data, but human auditors verify the final results. The founders' backgrounds include early Anthropic employment and leadership at METR, which performs similar evaluations for frontier labs, though METR has historically focused on performance rather than safety.
AIUC's certification could provide enterprises with a much-needed independent check on AI agent behavior, potentially accelerating adoption in regulated sectors like healthcare and finance. If widely adopted, it may reduce incidents of data leaks or harmful outputs, but could also create a false sense of security if standards fail to keep pace with rapidly evolving AI capabilities. Smaller firms might face cost barriers, while larger customers gain clearer risk assessments.