Overview
- The Open Secure AI Alliance released the Shared AI Findings Exchange, or SAFE, as a set of draft guidelines and tools that aim to collect confidential reports of AI incidents and near misses and convert them into evidence-based recommendations.
- The Linux Foundation has opened a public Request for Comments to manage review of the SAFE proposals and invite broader community input on governance and interoperability.
- More than 120 companies have joined the alliance and members are already publishing open-source tooling such as Nvidia’s Garak scanner, the NOOA audit harness, OpenShell runtime, Red Hat’s Asago, Amazon’s Cedar, and Microsoft’s red‑team tools to detect, sandbox, and test agent behavior.
- Several major AI platform providers, including OpenAI, Anthropic, and Google, are not currently listed as members, which could limit adoption because legal exposure, commercial competition, and disclosure rules may deter some participants.
- If widely adopted, SAFE could change how organizations respond to AI failures by sharing behavioral evidence and machine-readable controls that speed defensive updates and help regulators and firms map technical fixes to rules such as the EU AI Act.