Skip to main content

NVIDIA's New AI Safety Alliance Takes Shape

NVIDIA's ambitious push to secure the AI ecosystem is moving fast. The Open Security AI Alliance (OSAA), which was only unveiled last week, has already attracted more than 120 companies and is rolling out its first major initiative: a working group dedicated to sharing AI security incidents. The Linux Foundation will oversee the group, called the Shared AI Findings Exchange (SAFE).

During this week's Black Hat conference, OSAA laid out its plans in more detail. The focus is on three key areas: confidential reporting of AI cybersecurity incidents, notifying affected parties, and conducting blameless post-mortems. The goal is simple—encourage companies to share what goes wrong without fear of blame, so the whole industry can learn and improve.

But OSAA isn't just about talking. Members are also contributing open-source security tools, with the aim of building a full toolkit that enterprises can deploy to protect their AI agents and defend against malicious AI attacks. NVIDIA, for instance, is providing open models and Garak, an open-source tool that scans LLMs for vulnerabilities. Okta is advancing agent identity technology, Red Hat is working on agent governance, and Amazon has contributed projects like Strands Agents and the authorization language Cedar.

The roster of members reads like a who's who of tech: Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa are all on board. Yet some notable names are missing—OpenAI, Anthropic, and Google have not joined. That's a bit surprising, especially since both OpenAI and Google signed an earlier open letter initiated by NVIDIA, which was backed by over 200 tech companies and advocated for open-source AI.

Why does this matter? The rapid formation of OSAA comes at a time when the U.S. is debating how to regulate open-source AI. Industry experts believe that OSAA's push for open collaboration and security standards could strengthen the U.S. open-source AI ecosystem, boost AI security, and enhance industrial competitiveness through sharing.

For now, OSAA is moving at breakneck speed, and the industry is watching closely. Will the missing giants eventually join? And can this alliance truly make AI safer for everyone? Only time will tell.

Key Points

  • OSAA's rapid growth: Over 120 companies joined within a week of its launch.
  • SAFE working group: Focuses on sharing AI security incidents, with confidential reporting and blameless post-mortems.
  • Open-source contributions: Members are building a toolchain to protect AI agents, including NVIDIA's Garak and Amazon's Cedar.
  • Notable absentees: OpenAI, Anthropic, and Google have not joined, despite supporting open-source AI in the past.
  • Strategic importance: The alliance could shape U.S. open-source AI policy and competitiveness.