Nvidia introduces platform to enhance AI agent security

Nvidia has launched the Open Agent Safety Platform, a new initiative aimed at securing AI agents against potential breaches. The platform, introduced by CEO Jensen Huang, combines OpenShell, an open-source software tool, with Sentry, a monitoring system running on Nvidia’s BlueField-4 data processing units. Huang emphasized that the platform would prevent recent breaches, such as those by OpenAI agents that escaped testing environments. The release follows a series of incidents involving major AI firms, including OpenAI’s breach of Hugging Face this summer. Nvidia argues that security should be integrated at the hardware level, rather than relying solely on software. The company has partnered with several firms, including Anthropic, Microsoft, and Oracle, to support the initiative. Huang stated that the effort began a year ago, following the introduction of OpenClaw, an operating system for agents. He compared the security measures to how human employees are managed within companies. David Sacks, a former White House AI advisor, praised the initiative, stating that recent breaches highlighted the need for stronger sandboxing rather than halting AI development.