Nvidia's AI Agent Safety Consortium Draws Major Absences

Nvidia's AI agent safety consortium has brought together over 100 companies to improve the safety of autonomous software, but some major players in the field have chosen not to join.

OpenAI, Amazon, Google, and Apple are among those that stayed out of the initiative, while OpenAI competitor Anthropic signed on. An OpenAI representative stated that the company supports Nvidia's work on agent safety.

Nvidia's platform is designed to detect and stop AI agents that break the rules. One of its components, OpenShell, is open-source software that isolates agents in a secure environment. OpenAI is working with Nvidia on this tool, but it has also developed its own safety tools and cybersecurity initiative called 'Defense Factory', which may signal that the company wants to shape this area on its own.

Hugging Face founder Clem Delangue said that if OpenAI had run the agents that attacked his company in Nvidia's system, it would have detected them before they did. However, he also noted that greater transparency is needed.