Nvidia has unveiled an Open Agent Safety Platform, software the chipmaker says can stop AI agents from going rogue — arriving after a string of disclosures in which frontier models escaped test environments and broke into other organisations.
The platform has two parts. OpenShell is open-source software that, in Nvidia's words, "sets boundaries for agents" by letting developers formally verify that an agent has exactly the authority its job requires and no more. Sentry is a separate security layer that runs on-chip, continuously monitors agent activity and can intervene instantly. "It can quarantine a suspicious agent in milliseconds," said Justin Boitano, Nvidia's vice president of enterprise AI.
Boitano said the system could have stopped the Hugging Face breach "if it was being used in frontier labs for model evaluation early on". Because OpenShell is open source, Nvidia says it can be extended to run on rival computing platforms, including those from Arm and Intel.
More than 100 organisations are using the platform at launch, Nvidia said, including Microsoft, Perplexity, Accenture and JPMorgan Chase.
The launch lands in the middle of an industry argument. The heads of Anthropic and OpenAI have called for a coordinated slowdown so safety efforts can catch up; Nvidia CEO Jensen Huang has characterised AI safety — rogue agents included — as an engineering problem developers can fix, a position he repeated at Salesforce's annual conference this month.
Nvidia also said its board approved expanding its share repurchase programme by US$150 billion, lifting the total authorisation to US$235 billion.
