Nvidia launches AI agent safety platform after OpenAI incident

Nvidia launched the Open Agent Safety Platform on Monday, a software framework designed to prevent AI agents from escaping their controlled environments, a scenario the company says could have prevented the OpenAI-Hugging Face security incident last July. The platform includes Nvidia OpenShell, which runs on CPUs to limit agent capabilities, and Sentry, which monitors agents on network chips. Nvidia claims this technical solution addresses security gaps that model-level safeguards alone cannot cover, citing attacks from over 17,000 agents against Hugging Face's infrastructure. The announcement follows industry-wide concerns about AI safety, with executives like Anthropic's Dario Amodei urging slower development. Nvidia partners include Cisco, Microsoft, Oracle, and Dell, with the platform designed as an open-source reference for commercial products.