Everything you need to know about AI

Meta's Muse is a hit, and a privacy headache.

Meta's AI app reached 10 million US downloads in 40 days and keeps users better than its rivals. In the same week, shared Muse conversations exposed users' personal and business secrets, so Meta paused sharing, and about 80 WhatsApp help pages leaked hidden chatbot instructions. Meta also removed the word "train" from its commercial terms after AI backlash, but still won't commit to respecting robots.txt.

Sources TechCrunch ·Ars Technica ·Wired ·Ars Technica

Anthropic's draft safety bill draws fire as its compute squeeze grows.

Anthropic's proposed AI safety bill would only block a model that could kill 50 or more people, and critics note it stops federal agencies from halting a release without a judge's approval. Safety writers say the company has not grasped why the draft was troubling. Separately, Anthropic reportedly struggles to get enough compute from cloud partners that now see it as a competitor, ahead of a planned November IPO.

Sources TechCrunch ·LessWrong ·TechCrunch

Google open-sources TorchTPU to pull PyTorch users off Nvidia.

Google released TorchTPU, a native PyTorch back end for its TPUs, following Amazon in giving developers an easy path away from Nvidia GPUs. Hugging Face added native support on launch day. CUDA lock-in is now the main target for the big clouds' own chips.

Sources TechCrunch ·Hugging Face ·GitHub

AI-found bugs are overwhelming open-source maintainers.

curl will publish low and medium severity flaws openly from October 26 to survive an avalanche of reports, and PHP announced emergency security releases. Research cited by The Register shows a sharp rise in GitHub security fixes that break within 90 days, as AI speeds up attackers. On the defense side, Anthropic published work on using Claude to shrink a codebase's attack surface.

Sources The Register ·The Register ·Anthropic ·PHP

The US bans DeepSeek across federal agencies.

Federal civilian agencies must remove DeepSeek products and any third-party services built on its technology, citing national security. Critics say the order is vague and broad enough to be hard to apply in practice.

Sources TechCrunch ·The Register

Physical AI money keeps flowing.

Eka Robotics reached a $1.45B valuation on a robot arm that grasps varied objects with human-like smoothness, and Reliable Robotics raised $115M at over $1B for autonomous aviation. Gecko Robotics won a $300M US Navy contract, and Agility's new Digit humanoid is built to work 20 hours a day with fast-charging batteries.

Sources TechCrunch ·TechCrunch ·The Robot Report ·Robotics & Automation News