The Open Secure AI Alliance has formed a working group called the Shared AI Findings Exchange. It has already gathered more than 120 companies. The Linux Foundation manages the proposals coming out of this unit.
How it works
The group met at the Black Hat conference in Las Vegas last week. They are putting forward guidelines for reporting AI security incidents confidentially. The rules cover alerting affected parties and conducting blame-free analysis so everyone can learn from mistakes.
Members are also cataloguing open source technology. This collection could eventually become a way for companies to secure their AI agents or defend against rogue systems. Hugging Face is a member of the group and has previously dealt with an OpenAI model infiltration.
Nvidia offers a family of open models and a vulnerability scanner called Garak. Okta is building agent identity tech. Red Hat is working on agent governance. Amazon has contributed Strands Agents and the Cedar authorization language.
Who is in and who is out
Adobe, BlackRock, Cisco, Intel, Microsoft and Visa are now part of the alliance. Anthropic, OpenAI and Google are absent.
OpenAI and Google did sign the original open letter that created the group. Nvidia championed that letter, which asked the White House to support open source AI efforts. More than 200 tech companies signed it.
Anthropic has not joined the group or the letter. OpenAI and Google have released open weight models of their own. Google is known to support open source. It remains to be seen if they join the alliance as it gains momentum.
Speed and timing
The group is moving fast. News broke a couple of weeks ago that the Trump administration was considering banning Chinese open weight models. That news caused consternation in the industry and led to the open letter.
Whatever happens with those bans in the United States, the quick action by this heavyweight group is good for the US open AI ecosystem. Arcee, co-founder and chief technology officer of US open weight AI lab, says this is the best way to handle any threat from Chinese AI labs.
The industry group wrote that openness may be one of the most important paths to AI safety and security. They appear ready to put their effort and their technology into practice immediately.
What it means
People making AI tools now have a coordinated body to share security data and code. The focus is on practical fixes like reporting incidents and fixing vulnerabilities. This structure lets companies defend their systems without waiting for government mandates.
