Nvidia-Led Open Secure AI Alliance Moves to Enhance AI Security Through Open Source Efforts

The Open Secure AI Alliance (OSAA), a new industry group led by Nvidia, has formed a working group called Shared AI Findings Exchange (SAFE). This alliance, which includes over 120 companies, was established to improve AI security by developing proposals that are managed by the Linux Foundation and open for public comment.

• The group convened during the Black Hat conference in Las Vegas to deliberate on guidelines related to AI cybersecurity incidents, such as confidential reporting and blame-free analysis.
• OSAA members share open-source technology contributions that could lead to an industry-wide, open-source solution for securing AI tools and countering threats from rogue AI attackers. For instance:
  • Nvidia provides open models and an LLM vulnerability scanner called Garak.
  • Amazon has shared an agent building tool and an authorization language.
  • Other contributors include Okta with agent identity tech and Red Hat focusing on agent governance.
• The group features prominent members such as Adobe, Microsoft, Intel, and Visa, while notable companies like Anthropic, OpenAI, and Google are absent, despite their support for an initial open letter promoting open-source AI.
• This letter cautioned against U.S. legislation that might limit open-source collaborations, a reaction to the potential ban on Chinese AI models.

The initiative is seen as beneficial for the U.S. open AI ecosystem, with prominent voices like the CTO of Arcee advocating for openness as a strategy against perceived AI threats.