Open Secure AI Alliance launches SAFE framework for AI cybersecurity incident reporting

The Open Secure AI Alliance, founded by Nvidia and backed by more than 120 technology and security firms, has released the Shared AI Findings Exchange (SAFE) guidelines. Hosted by the Linux Foundation as a Request for Comments, SAFE proposes a confidential channel for organizations to report AI agent misbehaviour, security breaches and near‑misses, with analysis and evidence‑based recommendations provided by the alliance.

Key members contributing to the draft include Nvidia, Cisco, CrowdStrike, Hugging Face, Red Hat, Adobe, BlackRock, Capital One, Intel and Visa. Notably, OpenAI, Anthropic and Google are not participants. The alliance also unveiled a suite of open‑source tools for AI‑agent security and highlighted its rapid governance pace, delivering its first progress artifact within a week of launch.

The initiative aims to improve collective cyber defence by enabling shared learning across the AI stack, addressing the unique challenges of AI‑driven attacks that fall outside traditional vulnerability‑disclosure processes.