Nvidia doesn’t mess around: A week after open AI industry group formed, it’s already showing progress
TechCrunch AI Julie Bort ● Covered by 62 sources
Nvidia's week-old open AI safety group already has 120+ companies and real proposals on the table. Big names like OpenAI, Google, and Anthropic are sitting this one out despite backing the original open letter.
Nvidia doesn't waste time. A week after launching the Open Secure AI Alliance, the group already has a working committee, a cute acronym (SAFE, for Shared AI Findings Exchange), and actual draft proposals open for public comment, with the Linux Foundation handling the paperwork. Members hashed this out at Black Hat in Las Vegas, which feels fitting given the group's mission: figuring out how companies should confidentially report AI security incidents, warn affected parties, and run blameless postmortems so everyone learns from the next AI agent that goes sideways.
The proposals themselves are modest for now, nothing that rewrites the rulebook. But the real substance is in what members are bringing to the table. Nvidia is tossing in its open model family plus Garak, its open source LLM vulnerability scanner. Okta is building agent identity systems. Red Hat is tackling agent governance. Amazon contributed Strands Agents for building agents and Cedar, an authorization language. Add in more than 120 companies, including Adobe, Cisco, Microsoft, BlackRock, and Visa, and you start to see the outline of something bigger: an open toolkit for locking down AI agents before they get hijacked the way that OpenAI model reportedly slipped into Hugging Face.
What's missing is almost as telling as what's there. Anthropic never signed the original open letter and skipping the alliance tracks with its usual posture. But OpenAI and Google did sign that letter, the one urging the White House not to crack down on open source AI, and neither has joined the group they helped inspire. That's odd, especially since both have shipped open weight models and Google has long branded itself as open-source friendly. Maybe they're waiting to see if this thing has legs before committing.
The timing traces back to reports that the Trump administration was mulling a ban on Chinese open weight models, which spooked the industry enough to produce a 200-plus-company letter in a matter of days. Arcee's CTO has argued that openness, not bans, is the smartest defense against whatever threat Chinese AI labs actually pose. Nvidia and its allies seem to agree, and unlike most industry coalitions that spend a year drafting mission statements, this one shipped actual proposals before the ink dried.
My take
Credit where it's due: forming a 120-company alliance and producing draft security proposals within a week is genuinely fast for this industry, and it makes the letter-signers who stayed home look a little sheepish. OpenAI and Google can talk about openness all they want, but showing up costs nothing and their absence here reads as hedging, not principle. If the U.S. wants an open AI ecosystem that can out-compete Chinese labs on trust rather than trying to legislate them out of existence, this is the blueprint, and the companies dragging their feet should stop free-riding on the idea while skipping the work.
Read more about this at: TechCrunch AI
