Key facts
- The Open Secure AI Alliance (OSAA), formed a week ago and led by Nvidia, has already established a working group called SAFE.
- SAFE is proposing guidelines for confidential reporting and analysis of AI cybersecurity incidents.
- The alliance includes over 120 companies, with notable members like Microsoft, Intel, and Visa, but excludes OpenAI and Google.
- Members are contributing open-source technology aimed at securing AI agents and defending against AI threats.
- The group's rapid formation follows industry concern over potential U.S. bans on Chinese open-weight AI models.
The Open Secure AI Alliance (OSAA), an industry group initiated by Nvidia just a week ago and already boasting over 120 member companies, has rapidly established a working group named the Shared AI Findings Exchange (SAFE). This group is actively developing and proposing guidelines for the confidential reporting, notification, and blame-free analysis of AI cybersecurity incidents.
The SAFE group convened at the Black Hat conference in Las Vegas to formulate these initial proposals, which are now open for comment and being managed by The Linux Foundation. The guidelines aim to create a framework for learning from AI security breaches without assigning blame.
Beyond guidelines, OSAA members are contributing open-source technologies intended to help enterprises secure their AI agents and defend against potential threats from rogue AI. Examples include Nvidia's Garak LLM vulnerability scanner, Okta's work on agent identity, Red Hat's focus on agent governance, and Amazon's contributions of an agent building tool and an authorization language.
While prominent companies like Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa have joined the OSAA, notable absentees include Anthropic, OpenAI, and Google. This is particularly interesting as both OpenAI and Google had previously signed an open letter championed by Nvidia, urging White House support for open-source AI efforts.
The swift formation of the OSAA follows recent industry concerns over potential U.S. bans on Chinese open-weight AI models. Some in the U.S. AI ecosystem view the rapid action by this alliance as beneficial for the domestic open AI landscape and a potential counter to perceived threats from Chinese AI labs.