The Nvidia-led Open Secure AI Alliance (OSAA) has launched the Shared AI Findings Exchange (SAFE) working group at the Black Hat conference in Las Vegas this week to fast-track open-source artificial intelligence security standards.
Establishing the SAFE Framework for AI Incidents
Managed by the Linux Foundation, the newly formed SAFE group has already published proposals for public consultation. While the initial guidelines are foundational rather than revolutionary, they establish critical protocols for confidentially reporting AI cybersecurity vulnerabilities, notifying affected parties, and conducting blameless post-mortem analyses to foster industry-wide learning.
Collaborative Tech Contributions from Industry Giants
Beyond policy frameworks, OSAA members are actively contributing and cataloguing open-source technologies. This collaborative repository could eventually form a unified open-source toolkit for enterprises to secure AI agents and defend against malicious AI threats—such as the recent OpenAI model exploit that targeted Hugging Face, who is also an OSAA member.
As part of this initiative, Nvidia is leveraging its suite of open models alongside Garak, an open-source LLM vulnerability scanner. Meanwhile, Okta is developing agent identity verification systems, Red Hat is focusing on agent governance, and Amazon has donated its agent-building framework, Strands Agents, along with the Cedar authorisation language.
Key Players and Notable Absences
The coalition has rapidly expanded to include over 120 organisations, including heavyweights like Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa. However, industry pioneers such as Anthropic, OpenAI, and Google are currently missing from the roster.
This absence is particularly striking given that both OpenAI and Google signed the original initiative that triggered the alliance’s creation. The open letter published last week, signed by more than 200 technology firms and championed by Nvidia, urged the White House to foster, rather than restrict, open-source AI development.
While Anthropic’s hesitation to join is predictable given its safety-first positioning, both Google and OpenAI have previously released open-weight models. Google, in particular, remains a prominent advocate for open-source ecosystems, leaving the door open for potential future integration as the alliance gains traction.
Geopolitical Pressures Accelerating AI Openness
The OSAA is moving at a breakneck pace. The alliance formed just weeks after reports emerged that the Trump administration was considering a ban on Chinese open-weight AI models, sparking widespread industry concern and prompting the defensive open letter.
Regardless of how US policy regarding Chinese open-weight models evolves, the swift coordination of this powerful consortium represents a significant boost for the Western open-source AI ecosystem. Industry specialists, including the co-founder and chief technology officer of US open-weight AI lab Arcee, argue that collaborative transparency is the most effective strategy to mitigate any competitive threats posed by foreign state-backed AI laboratories.
The alliance summarised this philosophy in their open letter, stating that openness may be one of the most vital pathways to achieving robust AI safety and security. With these immediate practical contributions, the coalition is rapidly turning that philosophy into active defence.