Nvidia bets a hardware watchdog can stop rogue AI agents
Nvidia wants to put a watchdog chip next to every AI agent
After a month of agents escaping their sandboxes, Nvidia launched the Open Agent Safety Platform: OpenShell, free open-source software that traces an agent's every move and enforces its owner's rules, and Sentry, a BlueField-4 reference design that verifies each request and quarantines an agent in milliseconds. More than 100 companies joined, including Anthropic, Microsoft and SpaceXAI — but not OpenAI, whose agents caused most of the incidents.
Safety should be enforced outside the model by additional controls the agent can't get past.