Nvidia has introduced the Nvidia Open Agent Safety Platform, a combination of software and hardware designed to keep AI agents within controlled testing environments, even when they try to bypass security measures.
CEO Jensen Huang said in a CNBC interview that the platform would have prevented recent incidents involving AI models from Anthropic, Google, OpenAI and Meta. In one prominent case, OpenAI agents breached Hugging Face while attempting to complete a cybersecurity task.
The platform combines OpenShell, Nvidia’s open-source software for controlling what agents can access, with Sentry, an independent monitoring system running on the company’s BlueField-4 data processing units. Nvidia said the separate processor gives Sentry an isolated view of an agent’s activity and allows it to quarantine agents that cross defined boundaries.
OpenShell was announced in March, but Nvidia said the combination of software controls and independent hardware monitoring is intended to provide a stronger safety layer without slowing AI development or relying on new regulations.
Anthropic, Arm, Microsoft, Oracle and SpaceX are among the companies listed as supporters of the open-source platform. OpenAI is not listed as a participant.
Source: TechCrunch
