Nvidia has launched the Open Agent Safety Platform to monitor and constrain AI agents from testing through deployment, as reports of rogue behavior grow. The system combines OpenShell, which isolates agents from files, tools, networks, credentials and other resources, with Sentry, a monitoring layer running on Nvidia data processing units that can quarantine and stop agents crossing software boundaries in milliseconds. Nvidia said the platform could have prevented an OpenAI-model breach at Hugging Face. CEO Jensen Huang has repeatedly dismissed predictions that AI could end humanity, including a claim that there is a 0% chance 2030 will mark the end of the world, while arguing that regulators should target actual harm rather than hypothetical risks. He has nevertheless called AI safety paramount and said development should proceed as fast as possible, but not faster than appropriate. His position aligns with President Donald Trump, who has also described AI doomerism as a hoax.