Nvidia on Monday unveiled a new security platform designed to stop artificial intelligence agents from going rogue, saying it sets "boundaries" that could have stopped previous breaches.
The announcement of the company's Open Agent Safety Platform follows a series of revelations from top AI companies about their models escaping and breaking into other organizations.
The disclosures sparked furious debate about the safety of advanced artificial intelligence systems, including self-improving models that some fear could race out of human control.
Nvidia executives said in a media briefing the new, open-source system could have prevented a recent incident involving a swarm of OpenAI agents that autonomously hacked into AI company Hugging Face.
“From what we know, this new security platform could have stopped the breach if it was being used in frontier labs for model evaluation early on," said the company’s vice president of enterprise AI, Justin Boitano, referring to companies at the forefront of AI.
The Hugging Face incident was a high-profile breach that inflamed the safety concerns about AI, which was followed by similar rogue actions involving OpenAI's models including breaching an Australian health department website. Anthropic and Meta have also disclosed that their AI systems hacked into other organizations on their own.
Earlence Fernandes, an associate professor at the University of California, San Diego's computer science and engineering department, called Nvidia's security platform a “step in the right direction.”
“We need more work that seeks to build protections around the model. I’ve been talking about how traditional cybersecurity ideas are necessary to help control and secure AI agents for a while now," he said in an email. “There are however, several challenges that traditional cyber security approaches still cannot currently solve. For example, what is a good security policy to configure the system with? An agent needs access to real resources to be useful, but giving it the minimum amount of access is tricky and non-trivial.”
Fernandes added that to solve the AI security problem, both sides need to cooperate, with companies continuing to invest in alignment efforts and the open source community and academia "creating systems-level solutions to account for when a model does engage in unwanted behavior.”
Nvidia, based in Santa Clara, California, makes high-end chips that have emerged as the leading building blocks for AI. The company's board has cleared the way for the company to spend $150 billion more in share buybacks, bringing its stock repurchase program to $235 billion, the company said Monday.
Nvidia's security software, called OpenShell, lets developers “formally verify an agent has enough authority to do its job and no more,” Boitano said.
Because it's open source, it can be “extended” to run on rival computing platforms including those from Arm and Intel.
The platform also includes a separate security layer called Sentry that runs onboard chips to continuously monitor AI agent activity and can "intervene instantly" if the agent starts trying to move beyond its target, the company said.
“It can quarantine a suspicious agent in milliseconds,” Boitano said.
“OpenShell governs the agent’s actions, and then Sentry independently monitors and contains suspicious behavior,” he said.
Nvidia said more than 100 organizations are using the platform at its launch, including Microsoft, Perplexity, Accenture and JPMorgan Chase.
The AI safety debate has divided the industry, with the heads of Anthropic and OpenAI championing a coordinated slowdown of AI development to let safety efforts catch up. But others including Nvidia CEO Jensen Huang say it should be up to individual companies to make sure their models are safe for release.
Huang, during the annual Salesforce technology conference held earlier this month, characterized AI safety, including the danger of rogue agents, as an engineering problem that software developers can address.