The Open Secure AI Alliance is wasting little time. The week-old industry group, spearheaded by Nvidia, has already grown to over 120 companies and is now pushing its first security proposals into public discussion.
At the center of the early work is a new working group called the Shared AI Findings Exchange, or SAFE. Its focus is practical: create ways for AI cybersecurity incidents to be reported, shared, studied and turned into lessons without making blame the main event.
A Fast Start For An Open AI Security Effort
OSAA is moving at a pace that fits the current AI market. The group formed after an open letter published last week urged the White House to support open source AI efforts rather than restrict them. That letter was championed by Nvidia and signed by over 200 tech companies.
Now the alliance is trying to show that open AI can also mean organized AI security. SAFE has already developed proposals and put them forward for open comment. The Linux Foundation, which is a member of OSAA, is managing those proposals.
The work took shape as members gathered at the Black Hat conference, taking place this week in Las Vegas. That setting matters because the proposals are not aimed at abstract AI theory. They deal with the mechanics of cybersecurity coordination: how incidents are reported, how affected parties are notified and how the community can examine what happened afterward.
What SAFE Is Trying To Standardize
The first proposals are described as modest rather than dramatic. Still, the basics they cover are important if companies want shared AI security work to become repeatable instead of improvised.
Based on the source article, SAFE is looking at several connected areas:
- How to confidentially report AI cybersecurity incidents.
- How to alert organizations or people affected by those incidents.
- How to conduct blame-free analysis so others can learn from the event.
That combination points to a familiar cybersecurity challenge applied to AI systems. If companies do not trust the process, they may be less willing to share findings. If reports do not reach affected parties, the same weakness can keep spreading. If the follow-up becomes mainly about assigning fault, the learning value can shrink.
SAFE is not being presented as a finished answer. The proposals are open for comment, which means the group is still inviting feedback before the work hardens into any shared approach. For an alliance built around openness, that process is part of the message.
Open Source Tools Are Part Of The Plan
OSAA members are also contributing and cataloging pieces of open source technology that could be useful for AI security. The source article frames this as something that may eventually come together into an open source way for enterprises to secure AI agents or defend against rogue AI attackers.
The examples already span several parts of the AI security stack. Nvidia has pointed to a family of open models and an open source LLM vulnerability scanner called Garak. Okta is working on agent identity tech. Red Hat is working on agent governance.
Amazon has contributed Strands Agents, an open agent building tool, and Cedar, an authorization language. Hugging Face is also a member of the group, and the source article notes the example of an OpenAI model that infiltrated Hugging Face as the kind of rogue AI attacker scenario the ecosystem is trying to think about.
The importance of these contributions is not only that tools exist. It is that the alliance is trying to collect them in a common place and connect them to shared security practices. If that effort matures, enterprises could have a clearer starting point for securing AI agents instead of assembling every control from scratch.
Big Names, Notable Absences
The membership list already includes major companies across technology, finance and payments. Adobe, BlackRock, Cisco, Intel, Microsoft and Visa are among the names cited in the source article.
There are also conspicuous absences. Anthropic, OpenAI and Google are not listed as members. That is notable because both OpenAI and Google signed the original open letter that helped lead to this group, and both have released open weight models of their own. Google is also described in the source as generally known as a big supporter of open source.
The source article says Anthropic’s distance from the letter and the group so far is not a surprise. For OpenAI and Google, the question is whether they join later if the group builds momentum.
Why The Timing Matters
The alliance is emerging after news broke that the Trump administration was considering banning Chinese open weight models. According to the source article, that possibility caused concern in the industry and helped lead to the open letter.
The final shape of U.S. policy around Chinese open weight models remains outside what the source article establishes. What is clear from the article is that OSAA is trying to answer the policy moment with rapid collective action rather than waiting for the debate to settle.
Some in the open AI ecosystem argue that openness is the best way to meet any threat, whether real or imagined, that Chinese AI labs pose to the United States. The article specifically cites the co-founder and chief technology officer of U.S. open weight AI lab Arcee as part of that view.
“Openness may be one of the most important paths to AI safety and security,”
That line from the industry group’s letter captures the bet behind OSAA. The group is not only defending open AI in principle. Through SAFE and its growing catalog of open source contributions, it is trying to show that openness can be organized into security practice.