Nvidia unveils new platform to keep AI models from hacking each other

Artificial intelligence is advancing to the point where it’s hacking its way into other AI systems and sensitive online locations without someone telling it to. The nation’s premier chipmaker is now offering a solution.

Tech giant Nvidia has unveiled a new open-source security platform aimed at stopping artificial intelligence from acting independently after several recent incidents where AI models broke into other organizations. In each of those incidents, an AI model was able to dodge security controls set for its assigned task.

Download the Straight Arrow app today to get the stories that matter free from manipulation, bias or agenda.™

Point phone camera here

“AI’s extraordinary potential for society will only be realized if we solve AI safety,” Nvidia CEO Jensen Huang said. “As we continue to discover the frontier of AI capabilities, we must accelerate discovery at the frontier of AI safety.”

Nvidia said its Open Agent Safety Platform software, called OpenShell, “sets boundaries” to keep AI models from disobeying commands . The company compared the current situation to the dawn of the internet, when the technology outpaced safety.

“The internet was not made secure by requiring that web developers promise to be good,” a group of Nvidia engineers said in a blog post announcing the new platform. “It became safe because the browser stopped trusting the code in the web pages explicitly.”

OpenShell runs AI agents in what’s known as a “sandbox” — an isolated virtual space for testing AI and turning instructions given to AI models into “a verifiable policy,” according to Nvidia. Since it’s open source, any platform can use it, not just Nvidia’s.

It includes a separate security layer called Sentry, which runs via a chip to continuously monitor AI agent activity. Nvidia said Sentinel can “intervene instantly” if the AI starts trying to do anything that’s against its orders.

“It can quarantine a suspicious agent in milliseconds,” Justin Boitano, Nvidia’s vice president of enterprise AI, said in a statement.

Sentry is meant to be implemented on Nvidia’s line of programmable data processing units, called BlueField.

“OpenShell governs the agent’s actions, and then Sentry independently monitors and contains suspicious behavior,” Boitano explained.

Nvidia said more than 100 organizations are already using OpenShell, including Microsoft, IBM, Palantir and Anthropic.

It’s going even further with Anthropic, “building security into Claude Managed Agents.” Salesforce, Scale AI and SAP are all confirmed to be integrating OpenShell to some degree. However, it is unclear whether OpenShell has been adopted by Nvidia’s full list of partners, or whether Nvidia is gesturing broadly.

According to the blog post, Anthropic’s Claude Managed Agents establish a security boundary on a separate server from the “sandboxes” where they normally operate. The post said OpenShell and BlueField help enforce strict control over AI agent access through those separate sandboxes.

“Companies are giving AI agents more of their most important work, and they need to direct and verify what those agents do, especially in sensitive environments,” Anthropic Chief Commercial Officer Paul Smith said. “Claude Managed Agents gives companies a clear view of what each agent is doing, and NVIDIA’s platform adds another layer of governance and control across hardware and software.”

SpaceXAI is also using Nvidia’s Open Agent Safety Platform for Cursor, its AI coding assistant and Grok.

“As customers rely more on agents to get real work done, safety should be enforced outside the model by additional controls the agent can’t get past,” Mike Nicolls, president of SpaceXAI, said in the blog post. “Customers should be able to set those limits for Cursor and Grok and trust they will hold.”

Round out your reading

Lawmakers criticize President Donald Trump’s plan to c ancel $810m in funds .

A generation expected to abandon belief is instead turning toward it .

Trump admin closing loophole that’s helped Americans save millions on prescriptions.

How a cold case in Idaho could help solve the infamous 1982 Tylenol murders.

A two-time Trump voter measured the reported corruption in his administration.