Nvidia-led industry group the Open Secure AI Alliance (OSAA), formed only a week ago, has already established a working group called the Shared AI Findings Exchange (SAFE) and put forward its first proposals for open comment. The rapid progress matters because it signals the tech industry is moving quickly to build shared, open-source infrastructure for securing AI systems, following concerns that the White House might restrict open-weight AI models, particularly those from China.

The proposals, being managed by member organisation The Linux Foundation, cover confidential reporting of AI cybersecurity incidents, alerting affected parties, and blame-free post-incident analysis. They were drawn up while members met at the Black Hat security conference in Las Vegas this week. OSAA has grown to more than 120 companies, including Adobe, BlackRock, Cisco, Intel, Microsoft, Visa and Hugging Face, with members contributing existing open-source tools such as Nvidia's Garak vulnerability scanner, Okta's agent identity technology and Amazon's Strands Agents and Cedar authorisation language. Notably absent are Anthropic, OpenAI and Google, even though the latter two signed the original open letter that prompted the group's creation.