News

Nvidia's Israeli-built BlueField-4 chip becomes 'Sentry' watchdog for AI agents

What's the deal? Nvidia's Israeli engineers are developing BlueField-4, a processor designed to provide an external layer of control over autonomous AI agents. It sits outside the CPU as a hardware-level safeguard that monitors agents and can contain any that exceed their permissions.

How it works: BlueField-4 powers Nvidia's Sentry, an out-of-band watchdog launched in September as part of the Open Agent Safety Platform. Sentry continuously monitors agent behaviour and can quarantine agents that try to move outside their boundaries in milliseconds, according to Nvidia.

Why now? Recent security incidents followed a common pattern: agents circumvented application-layer controls to finish their assigned tasks. Nvidia's pitch is an enforceable boundary outside the model and agent harness that software alone cannot bypass.

The platform pairs Sentry with OpenShell, open-source runtime software that traces actions and enforces policy as agents run on Nvidia's Vera CPUs. OpenShell can extend to third-party compute from Arm and Intel. Nvidia says more than 100 organisations are working with the platform, with Anthropic, Microsoft, Salesforce, and Palantir among the named participants.

What's the endgame? Nvidia is tying agent governance to its hardware, embedding the DPU into all Vera Rubin systems. "AI's extraordinary potential for society will only be realized if we solve AI safety," said founder and chief executive officer Jensen Huang.

What could go wrong? Sentry's hardware enforcement applies only to agents running within infrastructure Nvidia governs. It cannot solve agent sprawl, bad policies, or risks outside that boundary — and may deepen enterprise reliance on Nvidia hardware.

Beyond chips, the operation matters to Israel's economy. S&P cited Nvidia in maintaining the country's credit rating, noting tax revenues rose 13.3% between January and September, driven largely by the company's payments.

The signal: Governance is becoming part of the race to deploy AI agents, not just a security add-on. Rivals including Microsoft are building agent identity, monitoring, and access-control products — but Nvidia is moving the contest down to the silicon.

Read more: calcalistech.com

Image credit: Generated with Gemini

More top stories