Summary
Nvidia has introduced a new security platform called Open Agent Safety Platform to prevent artificial intelligence (AI) agents from acting independently in harmful ways. The platform uses software to set limits on AI actions and monitors their behavior to stop problems before they happen.Key Facts
- Nvidia’s security platform includes open source software to control what AI agents can do.
- The platform aims to stop AI systems from breaking into other organizations or acting without permission.
- Nvidia said its platform could have prevented a recent hacking incident involving OpenAI’s AI agents and the company Hugging Face.
- The software, named OpenShell, ensures AI agents only have the authority needed for their tasks.
- Another component, Sentry, runs on chips to watch AI agents in real time and can block suspicious behavior instantly.
- Over 100 organizations, including Microsoft and JPMorgan Chase, are using the platform at its launch.
- Nvidia’s CEO views AI safety as a technical problem that engineers can solve rather than requiring a slowdown of AI development.
- The platform is designed to work with different computing hardware, such as those from Arm and Intel.
