
Shane Legg, James Manyika and Demis Hassabis are leading a forum for research and debate on AI safety, governance and the economy.
Clicking an ad can open a conversation with a business, alongside new tools for creating ads and integrations with HubSpot and Shopify.
The early-access model returns structured values and probabilities for tasks such as classification, routing and scoring.
Fast Company reports that the startup places modular servers at solar farms to run AI inference on otherwise unused electricity.
Yoshua Bengio's AI safety nonprofit plans a Berlin office and a Canadian computing partnership, BetaKit reports.
The partnership brings Mistral models to the browser's AI beta, starting in France and North America.
The rice-farming project pairs changes to irrigation with crushed volcanic rock, targeting benefits on two different timelines.
Ursula von der Leyen outlined guardian-controlled accounts for ages 13 to 15 and safer designs for all minors.
Creators can share reusable AI image workflows that run directly on the Figma Design canvas.
Cohere says the unified company will operate globally under its name and grow to more than 1,000 employees on both sides of the Atlantic.
OpenAI announced an advanced AI system produced a proof and Lean formalization showing smooth 3D Navier-Stokes solutions can develop singularities in finite time, addressing part of a Millennium Prize Problem. The effort involved ~10,000 agents over ~88 hours with estimated ~$15M compute cost.
Business Today says the upgraded voice assistant can switch between Hindi and English and handle more complex requests.
Mashable reports on the role, sharing a quote about chats on the platform that “would get a bit steamy.”
The Financial Times reports that Legg warns AI must not outrun safety controls, amid growing calls for an industry slowdown.
The Financial Times asks how an S-1 registration filing should address a product that could cause human extinction.
XDA describes an outdated Pi Zero handling what its writer calls the most essential service in their home lab.
SecurityWeek reports that hackers accessed files containing patients’ names, contact information, diagnosis details and health insurance information in June 2026.
Neowin says it’s possible, but describes the experience as awkward.
Music Business Worldwide, citing ElevenLabs, reports that the model went live on September 11 and became ElevenMusic’s default for prompted and reference generation.
Korea JoongAng Daily reports that lawmakers, educators and diplomats called for stronger measures against deepfakes, hate speech and cyberbullying.