From the archive
AI research tests new methods for training, evaluating and understanding artificial intelligence. Find news on machine learning papers, reasoning, model behaviour and experiments that challenge existing assumptions. Read what researchers measured, how they tested it and where the limitations lie, with links to the original work and discussion of what the findings could mean beyond a benchmark.
Lead story
Common Sense Media’s researchers say ChatGPT for Teens did not reliably alert parents in tests involving teenagers in crisis, while OpenAI disputes aspects of their methodology. The findings raise a practical question about whether the service’s safeguards work as intended when a young user may need help.
Read the report & discussion ↗Watch Desk
Donald Trump announced $2.4bn in AI compute pledges from 11 technology companies for the federal Genesis Mission, according to The AI Insider. The commitments are intended to give more than 15 federal agencies access to computing for scientific research, putting major AI suppliers directly behind a government research…
Watch Desk
Andreessen Horowitz says it is leading an investment in TypeSafe AI, the company behind Jev, a model designed to return structured decisions directly to software. The bet backs an alternative to the familiar chat-first approach: AI that supplies a usable result to code, rather than a paragraph for…
AWS AI Watch
AWS has brought metadata pre-filtering for Amazon S3 Vectors to its GovCloud US East and US West regions, and added a prefix-match operator for filters. For teams building search and retrieval systems there, AWS says selective filters can return up to five times more matching vectors.
Google DeepMind Watch
Google DeepMind’s Manish Gupta argues that AI safeguards should combine rigorous checks by developers with rules tailored to the settings where systems are used. His essay also points to safety work and partnerships the company says it is pursuing in India, where local language and domain needs are…
Allen Institute for AI Watch
Ai2 has replaced priority-based scheduling with GPU time budgets, fair-share allocation and rules for when workloads can be paused. The change is designed to give research teams a clearer share of scarce compute while keeping the cluster busy.
Watch Desk
The US Department of Energy, the National Institutes of Health and Biohub have signed an agreement to build an open data foundation for AI models of living systems. The plan pairs large-scale biological research with AI, laboratory automation and supercomputing, aiming to help researchers predict how cells respond…
GlobalFoundries Watch
GlobalFoundries has marked a major Dresden factory expansion and unveiled FDX Fusion, a planned chip platform for Physical AI. The investment could lift annual capacity at the site above one million 300mm wafers by the end of 2028, while the new platform is aimed at efficient computing in…
ElevenLabs Watch
Royal Mail used an AI-generated voice in its 2025 Christmas advert, and the daughter of late broadcaster Paul Vaughan says she recognised her father in it. Deadline reports that speech experts also found similarities, while Royal Mail says the voice was not designed to replicate anyone in particular.
OpenAI Watch
OpenAI says it banned Russian and Iranian accounts that used ChatGPT in influence campaigns across several countries. NPR’s account describes a striking split: Russian operatives mostly used the chatbot to report their work to superiors, while Iranian operatives used it to help produce material for publication.
Watch Desk
CIQ has made Fuzzball 4.3 generally available, a release designed to help organisations run open-weight AI models and agents on infrastructure they already operate. Its practical pitch is less hand-built integration: start a model, connect an agent and let the platform manage the GPU fleet underneath.
Firecrawl Watch
Firecrawl has published results comparing company-enrichment APIs, with Apollo scoring highest in its benchmark of four common fields. The post also describes how Firecrawl Alexandria lets AI agents call several enrichment providers through one connection, a practical pitch for teams tired of wiring up every API separately.
Watch Desk
Researchers at the Alan Turing Institute’s DARe centre and the University of Birmingham have built an AI model that identifies parts of satellites in orbital radar images. The work could help track what is happening to objects in increasingly busy orbits, though it remains research, not an operational…
Geordie Watch
AI safety is not just a question of which model an organisation approves. In a new article, Geordie.ai argues that an agent’s tools, access and surrounding software can shape what it does just as decisively, with practical consequences for how teams should secure and oversee deployed agents.
Watch Desk
Thread AI has agreed to work with the US Army on AI reasoning tools for research into Fire Control Systems, using the company’s Lemma platform. The project will help defence personnel search dense technical records, with the Army’s Armaments Center and Thread AI aiming to make answers traceable…
