Nvidia-led Open Secure AI Alliance shares early AI security proposals

In short: A week after forming, Nvidia’s Open Secure AI Alliance has published early proposals for how companies can share and learn from AI security incidents.

What happened

The Open Secure AI Alliance (OSAA) is a new industry group led by Nvidia. TechCrunch reports it has already grown to more than 120 member companies, including Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa.

The group has created a working group called the Shared AI Findings Exchange, or SAFE. SAFE has put out proposals for public feedback, and The Linux Foundation is managing the process.

The proposals were developed while members were at the Black Hat security conference in Las Vegas this week. The initial ideas focus on practical steps, like how to report AI-related security problems confidentially, how to alert people who may be affected, and how to do “blame-free” reviews so others can learn.

OSAA members are also sharing pieces of their open source software (software anyone can inspect and reuse, like a recipe that is published). Examples mentioned include Nvidia’s Garak tool, which can scan large language models for weaknesses, plus contributions from companies like Okta, Red Hat, and Amazon.

Notably, some major AI labs are not listed as members right now, including Anthropic, OpenAI, and Google. TechCrunch notes that OpenAI and Google signed the earlier open letter that helped spark this effort.

Why it matters

More companies are starting to use “AI agents,” which are AI systems that can take actions on their own, like a digital assistant that can click buttons and run tasks. If one of these systems goes wrong or gets abused, problems can spread fast. Shared reporting rules and shared tools can make it easier for companies to spot issues early and fix them.

Source: TechCrunch AI