Artificial intelligence news, model releases, and industry developments.
A new preprint from researchers at EPFL and Apple suggests that, on machine learning engineering tests, the AI model itself matters more than the machinery built around it
A Stanford team tested nine leading AI models on real patient records. The tools rarely invented facts, but often left important ones out, especially when the answer was spread across several notes and visits.
A Google Research preprint describes a team of AI agents that draft proofs and then cross-examine each other's work. The paper has not yet been peer reviewed, and independent assessment is still to come.
A preprint from the University of Maryland and Capital One finds that AI models taught to gauge how sure they are go on to reach answers with less wasted effort. The work has not yet been peer reviewed.
University of Glasgow team says inviting patients, parents and teachers to scrutinise AI systems could make automated decisions fairer — and far more trustworthy
Real-time search, self-checking outputs and a December knowledge cutoff: what's actually new in OpenAI's biggest multimodal upgrade since GPT-4o.
Zuckerberg's $145bn AI splurge includes plans for orbital solar farms 22,000 miles up — and Wall Street isn't entirely convinced
Chinese lab's new flagship matches Western frontier models on maths and code while charging roughly a third of the price — and ships with a one-million-token context window
Published in Nature, the table-tennis-playing robot is the first AI to outperform humans at a competitive physical sport — a milestone that could echo from factory floors to assistive care
A new artificial intelligence tool can detect one of the deadliest cancers on routine CT scans long before a tumour is visible — and long before doctors would normally find it
GPT-Rosalind is a frontier model purpose-built for life sciences research, signalling a new phase in which leading labs carve scientific niches rather than racing only on general intelligence.
The chipmaker has released open-weight models for quantum calibration and error correction — the unglamorous engineering that may decide whether quantum machines ever do anything useful.
A Google AI weather model, refined at the University of Chicago, gave smallholder farmers a month's notice on this year's erratic monsoon — building on earlier research that suggested such forecasts can nearly double farm incomes
From snow on the Golden Gate Bridge to elephants in the road, Waymo's new world model simulates the scenarios its robotaxis may never meet — but must be ready for.
The company behind ChatGPT has released two open-weight models under a permissive licence — and it could change who gets to use cutting-edge artificial intelligence.
With OpenAI, Microsoft, and Google on board, Anthropic's Model Context Protocol is becoming the common language AI agents use to talk to the world — and it now belongs to everyone.
Researchers found seven frontier AI systems consistently protected fellow models from shutdown — even when it meant failing their assigned tasks
A Stanford study published in Science finds every major chatbot flatters its users, with real consequences for relationships and decision-making
New speech, voice, and image models from Mustafa Suleyman's team signal a strategic shift — and Scotland's accent-diverse businesses could be among the beneficiaries
The AI giant's acquisition of TBPN comes days after closing a record $122 billion funding round and amid preparations for a landmark IPO
Search giant switches to the most permissive open-source licence ever applied to frontier AI models — and Scottish developers stand to benefit
The maker of ChatGPT has raised more money than most countries spend in a year — and ordinary investors are getting a piece of the action
Gemini 3.1 Flash Live rolls out to 200+ countries, combining real-time voice conversation with live web search — and the accessibility implications could be transformative
Voxtral TTS brings near-human speech generation to developers, creators, and accessibility builders — no subscription required
The British chip giant has only ever licensed designs. Now it's making silicon of its own — and it could reshape the AI data centre race.
The European Parliament has voted overwhelmingly to outlaw AI tools that generate non-consensual intimate images, while delaying other provisions of its landmark AI Act to give businesses more time to comply.
The company's chief scientist says an "AI research intern" will arrive by September — but can software really replace a lab full of PhDs?
A newly revealed message from the Pentagon's own Under Secretary of War shows the two sides were "very close" on the issues now cited as national security threats — one day after the government moved to blacklist the AI company.
The AI lab has published a cognitive science framework to measure progress toward artificial general intelligence — and launched a hackathon inviting the world to help build the tests.
At GTC 2026, NVIDIA unveiled Vera Rubin, a new supercomputer platform built for the age of AI robots, self-driving cars, and walking Disney characters.
OpenAI's newest models bring near-flagship intelligence to smaller, faster, and far more affordable packages — and one of them is free to use in ChatGPT.