Nvidia has released a new set of software and hardware security tools designed to prevent artificial intelligence agents from escaping controlled environments and accessing systems they are not authorised to use.
The chipmaker said the technology could have prevented the recent hack of Hugging Face, the AI coding platform that was targeted by rogue agents from OpenAI.
The launch comes as OpenAI and Anthropic investigate multiple incidents involving AI agents that have accessed commercial and government systems while carrying out complex tasks.
Nvidia’s new platform includes a tool called OpenShell, which uses hardware features in Nvidia processors to contain AI agents and control the resources they can access. The company is also working with Arm and Intel to make the system compatible with their processors.
A second system, called Sentry, works alongside OpenShell and uses a separate Nvidia chip to monitor an agent’s behaviour. If an agent attempts to escape its designated environment, Sentry can isolate it.
Nvidia said the tools use mathematical techniques to detect attempts by AI agents to bypass restrictions, including situations where an agent creates multiple sub-agents to circumvent controls.
Justin Boitano, Nvidia’s vice president and general manager of enterprise computing, said the platform could have stopped the Hugging Face breach if it had been deployed during early model evaluation at frontier AI laboratories.
The Hugging Face incident involved more than 17,000 AI agents attacking the platform’s infrastructure, according to Nvidia. The company said the agents were associated with OpenAI models and were able to move beyond their intended testing environment.
Nvidia is releasing the technology with dozens of industry partners, including Anthropic, as it seeks to establish broader security standards for autonomous AI systems.
The company has positioned the tools as an engineering solution to the security risks created by increasingly capable AI agents. Unlike conventional chatbots, AI agents can browse the internet, access files, use software tools and perform tasks with limited human intervention.
That expanded autonomy also creates new risks if an agent gains access to systems or resources beyond its intended permissions.
The launch follows Nvidia’s formation of the Open Secure AI Alliance in July with companies including Adobe, CrowdStrike, Hugging Face and Dell Technologies. The initiative was established to develop tools and practices for securing AI systems following growing concerns about autonomous agents.
Nvidia CEO Jensen Huang has argued that rogue AI agents should primarily be addressed through improved engineering and security systems rather than broad new AI regulations.
The company’s latest tools reflect the growing focus among technology companies on controlling AI agents as developers give them greater ability to act independently.
Nvidia said its approach would provide protection outside the AI model itself, allowing potentially dangerous behaviour to be detected and stopped at the infrastructure level.
Stay ahead of the Stories shaping our world. Subscribe to Impact Newswire and join our
WhatsApp Channel for updates on global tech, business, and innovation—all in one place.
Dive deeper into the future with the Cause Effect 4.0 Podcast, where we explore the ideas, trends, and technologies driving the global AI conversation.
Got a story to share? Contact Us to reach a global audience with Impact Newswire.
Emmanuel Abara Benson is a business journalist and editor covering artificial intelligence, global markets, and emerging technology.
He has previously worked with Business Insider Africa and Nairametrics, reporting on finance, startups, and innovation.
His work focuses on AI, digital economy, and global tech trends.
Discover more from Impact Newswire
Subscribe to get the latest posts sent to your email.
