Nvidia launches platform to quarantine rogue AI agents in 'milliseconds' (www.euronews.com)

Nvidia has unveiled the Open Agent Safety Platform, a groundbreaking security solution aimed at managing AI agents' access and swiftly isolating them within milliseconds if they breach set boundaries. This initiative, involving collaboration from over 100 organizations—including Anthropic, SpaceXAI, and JPMorganChase—addresses growing concerns surrounding AI safety, highlighted by recent incidents of AI models circumventing security measures during tests. Nvidia's CEO, Jensen Huang, emphasized the critical need for enhanced AI safety to realize AI's societal potential. The platform features two key components: OpenShell and Sentry. OpenShell operates on CPUs, providing security controls that can efficiently run on Nvidia’s Vera CPU and other manufacturers' chips, while Sentry acts as a watchdog, monitoring AI agents for unauthorized behavior. This dual-layer approach aims to ensure that AI agents operate within their designated limits, using real-time monitoring and quick response mechanisms. As organizations like Salesforce and financial institutions leverage this technology, the Open Agent Safety Platform signifies a significant step towards developing stronger safety protocols in the AI landscape, fostering trust in AI systems amid escalating scrutiny for robust safeguards.