The week-old Open Secure AI Alliance (OSAA), spearheaded by Nvidia and now boasting over 120 members, has made significant strides in developing guidelines for confidentially reporting AI cybersecurity incidents. While the proposals may seem basic, they mark a promising start towards a coordinated approach to AI security.

At the Black Hat conference in Las Vegas, members of the OSAA have also begun cataloguing open-source technologies that could be invaluable in securing AI systems. Companies like Nvidia, Okta and Red Hat are contributing their expertise, with Nvidia even offering an LLM vulnerability scanner called Garak.

The group's inclusion of big tech names such as Adobe and Microsoft is noteworthy, though some major players like OpenAI and Google have yet to join. Despite this, the OSAA is operating at a rapid pace and appears set to make a tangible impact on AI security.

As the debate over Chinese open-source AI models continues, this group's swift action could be seen as a positive move for the U.S. open AI ecosystem. The co-founder of Arcee, an American open weight AI lab, believes that openness might be crucial in safeguarding against any perceived threats from foreign labs.

It seems Nvidia and its allies are not just talking about open-source AI safety; they're actively working towards it. Only time will tell if this effort can bridge the gap left by absent giants like OpenAI.