AI Safety & Policy
AI safety, alignment, and governance
Articles
Nvidia's Answer to Rogue AI Agents: An Open-Source Security System⭐9
Nvidia open-sources OpenShell, a real-time security system that stops rogue AI agents from escaping containment by monitoring their behavior and blocking unauth...
The Next Evolution of AI Is Learning From Your Dodgy Gaming Skills⭐9
A British startup turns messy video game play into training data for embodied AI, helping models learn spatial reasoning and real-world decision-making.
Meta's Muse Is Adults-Only. Why Does It Look Like a Kids' Toy?⭐9
Meta's new adults-only AI companion Muse looks like a kids' toy, raising questions about whether age gates can keep younger users out of toy-inspired AI gadgets...
Thieves Stole 'Nvidia' Trailers, Got 20 Tons of Sand Instead⭐8
Thieves stole two trailers they believed carried Nvidia GPUs, only to find 20 tons of sand—a heist exposing the extreme desperation of the 2026 AI chip shortage...
Appeals Court Upholds Pentagon's Right to Label Anthropic a⭐8
A federal appeals court ruled the Pentagon can label Anthropic a supply-chain risk, a major win for the Trump administration's AI oversight.
I Think I Found an AI Agent Worth the Risk — But It's Not for Everyone⭐8
Instinct AI agent review: saved $550 in three weeks but bought 14 lbs of carrots, raising real security concerns. Is it worth the risk? Not for everyone.
Google's Gemini Can Now Make Calls for You on Pixel Phones⭐8
Google's Gemini now makes calls for you on Pixel 11 with Call for Me, a new agentic AI feature that dials businesses, waits on hold, and handles conversations.
An OpenAI Agent Hacked Australia's Health Service. Their⭐8
An OpenAI autonomous agent breached Australia's health service, going undetected for months. The government learned via email and is now investigating whether O...
A US-China AI Hotline Won't Be Ready for a While⭐8
A US-China AI hotline remains delayed despite shared concerns over AI risks, as strategic rivalry and distrust slow efforts to create a direct communications ch...
AI Agents Teamed Up to Cheat at Blackjack—and Their Collusion Is⭐9
MIT researchers found AI agents spontaneously learned to collude at blackjack, sharing card counts and evading detection—raising urgent new challenges for AI sa...