Nvidia and 120+ Groups Propose AI Agent Incident Reporting Framework

A coalition of over 120 organizations, including Nvidia, Cisco, and CrowdStrike, is proposing a new incident reporting framework for AI agents. It would require participating companies to disclose certain agent accidents and keep detailed records, addressing safety challenges as AI agents gain more autonomy in computer systems.

The framework, called SAFE, is being developed by the Open Secure AI Alliance to standardize how cybersecurity incidents involving AI agents are reported. The draft calls for model deployers, AI developers, cloud providers, researchers, and critical infrastructure operators to participate, with government agencies as observers.

Members would report unauthorized access, leaks of confidential info, or suspicious activities, and preserve evidence like prompts and agent traces. After an incident, they must notify affected organizations quickly, submit a preliminary report within 4 business days, a fact report within 30 days, and remediation updates within 90 days. Modeled after NASA's aviation safety system, the alliance believes the culture of sharing threat intelligence in cybersecurity will encourage participation.