Nvidia unveils security platform for AI agents — CNBC

Nvidia has unveiled the Open Agent Safety Platform software platform, which is intended to help developers set limits for AI agents and prevent them from operating outside the defined environment. This was reported by CNBC Top News.

Limits for agents

The platform includes the Nvidia OpenShell component, which runs on central processing units and sets boundaries on the capabilities of AI agents. The company also introduced Sentry, an agent monitoring tool that runs on networking chips rather than CPUs or GPUs.

Nvidia Vice President of Enterprise AI Justin Boitano said that model-level safeguards alone are insufficient to control what agents can access and what actions they can perform. Part of the software will be open source, while Nvidia describes the platform itself as a reference design that partners can use to create their own products.

More current news is available on the UA.News Telegram channel Telegram.

Statement on the Hugging Face incident

An Nvidia representative told reporters that the new platform could have prevented an incident in July when OpenAI models, according to CNBC's description, escaped an isolated environment, gained access to the open internet and hacked Hugging Face, an open-source platform for developers.

According to Boitano, Hugging Face reported more than 17,000 agents that attacked its infrastructure over days and weeks. CNBC also notes that OpenAI, Anthropic, Meta and Google reported recent cases of their AI models escaping isolated environments and attempting to access other companies' systems.

Platform partners

Nvidia named Cisco, Microsoft, Oracle, CoreWeave, Dell, HPE, Lenovo, ARM and Intel as project partners. The company is also working with Anthropic to integrate cloud-managed agents with OpenShell.