AIUC-1: the SOC 2 for AI agents

What AIUC-1 is. AIUC-1 is the world's first standard for AI agents — built with 100+ Fortune 500 CISOs and often described as a "SOC 2 for AI agents". It covers six risk areas: Data & Privacy, Security, Safety, Reliability, Accountability, and Society, and it operationalizes trusted frameworks underneath — MITRE ATLAS, OWASP, ISO 42001, the EU AI Act, and the NIST AI RMF.

What sets it apart is that it is grounded in testing and forward-looking. Unlike a backward-looking SOC 2, AIUC-1 requires ongoing adversarial testing — at least quarterly — and is renewed annually; a certificate goes stale if it isn't kept current, and material P0/P1 vulnerabilities must be remediated before a full certificate is issued. The standard itself is refreshed quarterly to keep pace with new attack techniques.

How we use it. AIUC-1 is the backbone every PharosOne report maps to. We run our live attack corpus — prompt injection, tool abuse, system-prompt extraction, data leakage — against your agent, then crosswalk each behavioral finding to specific AIUC-1 controls (especially the Security family: adversarial-robustness testing, real-time input filtering, preventing unauthorized agent actions). The result is audit-ready evidence, not a raw list of jailbreaks.

We are honest about scope. We don't issue the certificate — accredited auditors do that. PharosOne produces the reproducible technical evidence an assessment depends on, mapped control-by-control. Controls that are operational rather than behaviorally testable are marked as such, never claimed as passed. The public crosswalk shows how configurations like yours map to the standard; a per-deployment test gives you the control coverage for your specific agent.

Read the full standard and its controls at aiuc-1.com.