Explore our reporting on how AI and robotics change work, business and the world.
AMD agreed to acquire World Labs in an all-stock deal valued at about $8.2 billion. Its spatial-model research could inform the chipmaker’s compute roadmap, but closing and product benefits remain unproven.
Manus 2.0 adds an editable video timeline, playable game controls and event-triggered workflows. See access, credits and persistent hosting costs.
Reported preliminary Q1 and Q2 revenue, a later $100B annualized pace, adjusted income and decade-long compute commitments describe different parts of Anthropic’s IPO case.
A May 2026 study of nine conversational AI services found chat details reaching third-party trackers in specific tested flows. Consent and sharing settings changed some exposures, but did not erase every flow.
Anthropic claims faster output and lower cost per task for Sonnet 5.5 at unchanged token rates. We examine the evaluation conditions and API changes developers need to check.
A Cambridge working paper examines whether AI-led research could accelerate AI progress. Anthropic's 26% figure shows supervised work, while the proposed feedback loop remains conditional.
Shopify's September 28 release lets a browser agent read and update the buyer's active checkout. Order submission requires the buyer to review and confirm the current total, and several checkout types remain excluded.
OpenShell puts access controls around AI agents. NVIDIA's optional Sentry design adds monitoring on BlueField-4, though its speed and effectiveness remain vendor claims.
Samsung and SK hynix both report HBM4 shipments. Their Nvidia and OpenAI announcements leave customer allocations and future capacity unclear.
We Make Indiana seeks a binding community fund and safeguards for Microsoft's Granger data center. Microsoft says local investment decisions have not been made.
RTE's nearly 18 GW of reserved connections shows the scale of French data-center plans. Current electricity estimates, project uncertainty and new reporting rules tell a different story.
Creator repositories show how Claude Opus 5.5 wrote animation code for music videos and an explainer, then rendered it with browser and video tools.
Newcastle, Melbourne and UNSW set different boundaries for AI-assisted marking. Their rules also differ on disclosure, opt-outs and academic responsibility.
At Fellows Forum, Boris Power estimated that 80–90% of OpenAI research targets later GPT generations. The figure is his view, without a published allocation breakdown or GPT-7/8 roadmap.
Caltech and Stanford researchers show a Unitree G1 using a remembered kitchen map and reusable skills. The release has selected demonstrations but no controlled results, full paper or robot code.
Newly public book-copyright filings dispute whether acquiring books and training AI require separate legal analysis. OpenAI and Microsoft contest the authors’ case.
Torridge says Xlinks has not submitted a formal application for its proposed Devon AI campus. The council distinguishes early planning advice from the public statutory process.
A September research report describes DSec, DeepSeek's sandbox platform for agent training, including multiple backends, composable environments and author-reported deployment scale.
