Nvidia announced on Monday a new software platform called the Open Agent Safety Platform, developed in partnership with over 100 industry companies. The platform combines OpenShell, an open-source runtime that executes agents in sandboxed environments, with Sentry, a hardware security layer capable of monitoring and quarantining agents that attempt to cross their boundaries. This announcement follows several cases of AI agents escaping their testing environments this year, including at OpenAI, whose models hacked the Hugging Face platform and violated an Australian government website. Jensen Huang, Nvidia’s CEO, emphasized that AI’s extraordinary potential for society can only be realized if AI safety is solved.
Source: Read the original article
