According to a report by Julie Bort in TechCrunch, the Open Secure AI Alliance (OSAA)—a new industry group formed just a week ago and spearheaded by technology giant Nvidia—is already showing significant and rapid progress. In this remarkably short period, the group has grown to include over 120 companies. It has announced the establishment of a uniquely named working group: the Shared AI Findings Exchange, or SAFE for short. This working group is already presenting its initial proposals for public comment, which are being officially managed by the Linux Foundation, a member of the alliance. The group's members developed these guidelines while gathering at the very center of the cybersecurity world: the Black Hat conference held this week in Las Vegas.
SAFE Group and the Initial Security Guidelines
The guidelines introduced by the SAFE working group are not particularly earth-shattering or revolutionary at this stage, as noted in the report, but they lay important foundations for collaboration in the field of artificial intelligence security. The submitted proposals focus on several key areas of cybersecurity within the AI domain. Among other things, the guidelines cover how to confidentially and securely report AI-related cybersecurity incidents, how to alert the affected and impacted parties, and how to conduct a blame-free analysis of these incidents. The goal of this non-punitive approach is to enable all industry players to learn from these events and improve their systems for the future.
Alongside the publication of these guidelines, member companies of the OSAA are actively contributing and cataloging portions of their open-source technologies that could be useful to other members of the alliance and the industry at large. As often happens with industry organizations of this nature, these collaborative efforts may eventually coalesce into a comprehensive open-source framework or tool. This would allow enterprises to secure their AI agents or defend themselves against rogue AI attackers. The report cites as an example of such a threat the OpenAI model that infiltrated the Hugging Face platform—notably, Hugging Face itself is a member of the newly formed OSAA.
Technology Contributions from Alliance Members
Key member companies of the OSAA have already begun offering practical open-source tools and solutions to advance the group's security goals. Nvidia, for example, noted that it offers an entire family of open models, alongside Garak, a dedicated open-source vulnerability scanner specifically designed for large language models (LLMs). The security firm Okta is working on developing technology designed for agent identity verification (agent identity tech). Red Hat is focusing its efforts on developing solutions for agent governance and management (agent governance).
Furthermore, technology giant Amazon has contributed two significant tools to the collaborative effort: an open tool for building AI agents known as Strands Agents, as well as an authorization and access language called Cedar. These contributions represent only a fraction of the many technologies being brought to the table by alliance members, with the ultimate goal of establishing a secure, standardized environment for developing and deploying open AI tools across the entire industry.
Alliance Composition, Prominent Members, and Notable Absences
The OSAA has successfully attracted a long list of prominent names from the technology and business worlds. Among the companies that have already officially joined the alliance as members are Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa. However, despite the widespread response, there are several highly notable absences from the roster, including Anthropic, OpenAI, and Google, which have not yet joined as formal members of the alliance.
The absence of OpenAI and Google is of particular interest because both of these companies signed the original open letter that sparked the creation of this group. Published last week, the open letter urged the White House administration to support open-source AI initiatives rather than suppressing or crushing them. This letter was championed and led by Nvidia, garnering signatures from more than 200 different technology companies. While Anthropic’s cold shoulder toward both the letter and the industry group so far does not come as a surprise, both OpenAI and Google have previously released their own open-weight models. Furthermore, Google is generally recognized as a major supporter of open source. It will be interesting to observe whether these companies choose to join the alliance as the group gains momentum and industry influence.
Political Background and Impact on the U.S. Ecosystem
The establishment of the alliance and its rapid activity are unfolding at a breakneck pace, mirroring the speed of development in the AI industry itself. Only two weeks have passed since reports surfaced suggesting that the Trump administration was considering a ban on open-weight AI models originating from China. These reports caused deep concern and anxiety within the industry, which ultimately led to the drafting and publication of the open letter to the White House.
Regardless of what ultimately happens with Chinese open-weight models in the United States, the fast and decisive action of this heavyweight coalition of companies is viewed as a positive development for the open AI ecosystem in the U.S. Various figures in this ecosystem, such as the co-founder and Chief Technology Officer (CTO) of Arcee—a U.S.-based open-weight AI model development lab—argue that this open approach is the best way to counter any threat, whether real or imagined, posed to the United States by Chinese AI labs. As the industry group members wrote in their joint letter: "Openness may be one of the most important paths to AI safety and security." It appears that the alliance members are ready to prove this immediately by putting their effort and practical technology into action on the ground.