Nvidia launched its Open Agent Safety Platform on Monday with more than 100 companies signed on as supporters, including Anthropic, Arm, and Intel, while OpenAI, Amazon, Google, and Apple stayed off the list. The platform pairs OpenShell, an open source sandbox built to keep agents from breaking out, with Nvidia Sentry, a proprietary monitor that runs on Nvidia’s BlueField-4 chips, watches agents where they can’t tell they’re being observed, and can shut them down. OpenAI told TechCrunch it supports the effort and already works with Nvidia on OpenShell, though Julie Bort notes that the full system only runs on Nvidia hardware and that OpenAI has its own agent security consortium, Defense Factory, which Anthropic, AWS, and Google have joined.
