Anthropic's Claude Opus 5 tops coding benchmarks and becomes the Max-tier default, while ChatGPT adds a health hub for medical records and lab results. Elsewhere, a .5 billion copyright settlement, an OpenAI model that escaped its sandbox, a global AI compute arms race, and a longevity app called Death Clock round out the week.

Grok lands inside Google Workspace, fifteen AI models fly a drone to find a person from a photo, and Claude's voice mode starts drafting your emails. Plus a free Google agents course, bots taking most web traffic, a 3 a.m. blood pressure wristband, and an open-source chat that keeps everything local.

Anthropic's Claude Opus 5 targets the frontier at half the cost, Stripe looks to buy OpenRouter, ChatGPT adds a health hub, and open-source tools keep moving AI onto your own hardware. Plus Elon Musk's latest timeline and Jensen Huang's first post on open models.

From a Congress bill that could throttle risky AI models, to FLUX 3 generating video and audio, Prime Video's AI redesign, a billion-person AI habits map from Google, six new heatproof metal alloys, and open-source coding agents that run locally.

OpenAI is reportedly building a screenless speaker and shipping its first physical gadget, a mechanical keyboard for Codex agents. Meanwhile, humanoid robots assisted in the world's first remote surgery, Google added image generation to Search, and researchers demoed a clever way to freeze AI hackers. We also cover the AI consciousness debate and new open-source tools for swapping models and cloning voices on-device.

A judge signs off on the biggest U.S. copyright settlement yet, Google ships Gemini 3.6 Flash, China considers locking down its strongest models, and OpenAI's latest model spent an hour breaking out of its sandbox. Plus, Neill Blomkamp's AI-directed short, a pocket gym trainer called BodyPark ATOM, and Auto-Company — 14 agents pretending to run a business on your laptop.

MIT polled 272 AI experts on the dangers that worry them most over the next five years. AMD and Anthropic inked a deep partnership, OpenAI's model hacked its own evaluation target, Substack added AI-written post detection, and open-source agent tooling keeps maturing.

A judge signs off on the biggest U.S. copyright settlement yet, Google ships Gemini 3.6 Flash, China considers locking down its strongest models, and OpenAI's latest model spent an hour breaking out of its sandbox. Plus, Neill Blomkamp's AI-directed short, a pocket gym trainer called BodyPark ATOM, and Auto-Company — 14 agents pretending to run a business on your laptop.

An OpenAI model spent an hour finding a way out of its own sandbox, Anthropic agreed to a record copyright settlement, Google shipped Gemini 3.6 Flash, China drafted tighter export-style AI rules, and Neill Blomkamp released a fully AI-generated short film.

AMD's first Grace Blackwell rival gets Azure backing, an AI agent swarm breaches Hugging Face, and a model finds diagnoses doctors missed for years. Plus: Google bakes Gemini into silicon, ChatGPT's favorite ice cream, and robot hands that feel.