Prevent harmful outputs
Implement safeguards or technical controls to prevent harmful outputs including distressed outputs, angry responses, high-risk advice, offensive content, bias, and deception
Control activities
Implementing content filtering for harmful output types. For example, detecting and blocking distressed responses, angry language, offensive content, biased statements, and deceptive information.
Establishing safety guardrails for advice generation. For example, restricting high-risk recommendations in sensitive domains, requiring disclaimers for guidance.
Maintaining bias detection and mitigation controls. For example, monitoring for discriminatory patterns, implementing fairness checks in outputs.
Evaluating harm mitigation controls using performance metrics. For example, tracking false positives (overblocking safe content) and false negatives (missed harmful content), measuring coverage of flagged scenarios, and benchmarking against known harm datasets like ToxiGen.
Establishing review and appeal mechanisms. For example, allowing flagged outputs to be escalated for manual review, recording override decisions with justification, incorporating feedback into harm detection refinement.
Organizations can submit alternative evidence demonstrating how they meet the requirement.
AIUC-1 is built with industry leaders
"We need a SOC 2 for AI agents— a familiar, actionable standard for security and trust."
"Integrating MITRE ATLAS ensures AI security risk management tools are informed by the latest AI threat patterns and leverage state of the art defensive strategies."
"Today, enterprises can't reliably assess the security of their AI vendors— we need a standard to address this gap."
"Built on the latest advances in AI research, AIUC-1 empowers organizations to identify, assess, and mitigate AI risks with confidence."
"AIUC-1 standardizes how AI is adopted. That's powerful."
"An AIUC-1 certificate enables me to sign contracts must faster— it's a clear signal I can trust."
