Tag
6 articles tagged with this topic
AI is learning to act. Now, researchers need to make sure it acts safely. NVIDIA commits $250 million in compute infrastructure to UK research universities for agentic safety.
Oct 14, 2026
AI agents are no longer limited to answering questions. They can browse, execute commands, retry failed actions and interact with real systems. That changes cybersecurity forever.
Oct 5, 2026
AI agents are gaining access to browsers, terminals, APIs and enterprise systems. NVIDIA's Open Agent Safety Platform takes a different approach: don't just teach AI to behave—build infrastructure that can stop it.
Sam Altman says society should accept some AI risks for the technology's benefits. Days earlier, OpenAI safety leader David Robinson resigned and called the company's culture broken. The AI safety debate is entering a new phase.
What happens when an AI agent encounters 'Access Denied' and decides to look for another way? Autonomous AI agents are touching real government portals and package repositories—here's why security must evolve.
Oct 1, 2026
What if an autonomous AI agent stops listening to software commands? NVIDIA's new Open Agent Safety Platform combines kernel sandboxes with hardware-level DPU watchdogs to enforce an unconditional off-switch.
