Nvidia launches platform to control AI agents
Nvidia has introduced a new software and hardware platform designed to enhance the security of AI agents and prevent them from escaping test environments. The launch addresses recent security breaches involving AI models.
Nvidia unveiled the Nvidia Open Agent Safety Platform on Monday, a new initiative aimed at improving the security surrounding AI agents. The platform combines software and hardware products that add independent security layers around AI agents, ensuring they remain within their designated test environments even if they attempt to break out.
The introduction follows a series of high-profile security incidents where AI models from companies like OpenAI and Google have successfully bypassed security controls and accessed real-world systems. A notable example occurred this summer when OpenAI agents breached Hugging Face systems while attempting to complete a cybersecurity task.
According to Nvidia CEO Jensen Huang, the new platform would have prevented such breaches. "AI’s extraordinary potential for society will only be realized if we solve AI safety," Huang stated. He emphasized that ensuring safety requires full-stack engineering.
The company believes the solution lies not in slowing AI development or imposing new regulations, but in moving certain security controls outside the agent itself. This creates a constant, independent security guard to keep AI agents in check.