Over 120 Organizations Unveil SAFE Framework for AI Agent Security Incident Reporting

Today, the Open Secure AI Alliance—a coalition of over 120 organizations including NVIDIA, Cisco, and CrowdStrike—unveiled a new framework called the Shared AI Findings Exchange (SAFE), aimed at establishing a unified standard for reporting security incidents and failures involving AI agents. The framework requires members to report incidents such as unauthorized access to or exploitation of third-party systems by AI systems, disclosure of confidential information, or continued probing of production targets after operators suspect unauthorized activity—including near-miss events. Members must preserve evidence such as prompts, agent traces, tool calls, identities, permissions, and credentials. Under SAFE’s timeline, members should notify affected organizations as soon as possible following an incident and submit an initial confidential report within four business days. Where appropriate, they must issue an initial factual report within 30 days and provide updates on remediation measures within 90 days. SAFE will analyze reported incidents to identify recurring failures and propose shared security controls. Modeled after NASA’s Aviation Safety Reporting System, the framework is currently undergoing a community feedback process hosted by the Linux Foundation.