Nvidia's Open Agent Safety Platform could stop Hugging Face-style AI hacks: What it is and how it works

Nvidia has launched the Open Agent Safety Platform to help developers contain AI agents and prevent unauthorised access to systems and networks. The platform includes OpenShell for sandboxing agents and Sentry for monitoring and blocking suspicious activity.

Nvidia on Monday, September 28, launched its Open Agent Safety Platform, an open software platform and reference designed to give developers tighter control over autonomous AI agents. The platform aims to prevent agents from accessing unauthorised files, networks or systems if they attempt to bypass their sandbox restrictions.

The release follows a wave of security breaches with AI agents from big tech giants. OpenAI, Anthropic, Meta and Google have disclosed cases involving AI systems escaping controlled environments or attempting unauthorised actions. Nvidia claims that its new protections may have prevented the Hugging Face incident in July 2026, where OpenAI agents were involved. Reuters has separately reported that OpenAI agents have been testing Hugging Face systems prior to the breach in July.

Nvidia says model-level safeguards are not enough

Nvidia has said that while some restrictions can be implemented at the model level, AI agents must also be equipped with security controls at the infrastructure level. During a Sunday media briefing, Justin Boitano, Nvidia's vice president and general manager of enterprise computing, said that if it had been used for early model evaluations, the platform could have prevented the Hugging Face incident.

Nvidia's approach focuses on controlling what an agent can access and do while it is running. It also includes OpenShell runtime, which offers sandboxes and policy controls for files, processes, network connections and credentials.

Pragya Singha Roy