Nvidia-led Open Secure AI Alliance (OSAA) formed its first working group, SAFE (the Shared AI Findings Exchange), a week after its launch and released a proposal on responding to AI security incidents.

TechCrunch reported on Aug. 4 that the number of OSAA participating companies has already surpassed 120.

The working group prepared the proposal on the sidelines of the Black Hat security conference held in Las Vegas, and the Linux Foundation took charge of collecting public comments and managing the proposal.

The proposal includes a way to confidentially report AI cybersecurity incidents, procedures to notify affected parties and a post-incident analysis system focused on learning rather than assigning blame. It does not immediately introduce major technical changes, but focuses on first establishing common industry response rules.

OSAA participants are also cataloguing open-source technologies that can be used for AI security.

Company contributions are also continuing. Nvidia is providing several open models and an open-source large language model vulnerability scanner called "Garakk". Okta is providing agent identity technology, and Red Hat is developing agent governance. Amazon released an agent-building tool called "Strands Agent" and an authorization language called "Cedar".

The alliance includes Adobe, BlackRock, Cisco, Intel, Microsoft and Visa. Anthropic, OpenAI and Google have not yet joined.

OpenAI and Google signed a letter to the White House last week urging it to support and not suppress open-source AI. Nvidia led the letter, and more than 200 technology companies put their names on it.