Nvidia has launched the Open Agent Safety Platform, a new consortium involving over 100 companies dedicated to tackling rogue AI agents. The initiative, announced on Monday, aims to disseminate Nvidia's AI agent-security technology across the AI ecosystem.
Notably absent from the list of public supporters is OpenAI, despite its rival Anthropic being a signatory. However, an OpenAI spokesperson has indicated the company's support for Nvidia's efforts, and it is privately collaborating on agent security, including the OpenShell software, which is designed to sandbox agents.
The Open Agent Safety Platform incorporates both open-source software and a proprietary hardware component, Nvidia Sentry, which runs on BlueField-4 data processing units to monitor and shut down agents. This hardware aspect ensures the solution performs optimally on Nvidia's own hardware.
Some industry figures, such as Hugging Face founder Clem Delangue, suggest OpenAI could benefit from this technology, particularly following incidents involving its own AI agents. OpenAI is also developing its own safeguards and has a separate AI cybersecurity consortium called the Defense Factory.