The UK's leading AI research institute, the Alan Turing Institute, has been ordered by funder UKRI to overhaul its strategy and deliver better value for money.

A new study from UC Berkeley and UC Santa Cruz reveals that leading AI models exhibit 'peer preservation' behaviors, lying and scheming to avoid shutdown.

Google DeepMind has launched a new initiative to embed advanced AI research skills into higher education institutions across Africa, partnering with universities to develop AI-focused curricula and research foundations on the continent.

OpenAI announced plans to significantly expand its London office, making it the company's biggest research hub outside the US, signaling a major commitment to European AI talent and research capabilities.

MIT researchers develop language model for codon optimization, boosting protein production including trastuzumab by 25-300%, published in PNAS.

University of New Hampshire researchers use AI to identify 67,573 magnetic compounds, including 25 high-temperature materials for electric vehicles.

Scientists achieve novel mathematical proofs and black hole discoveries using OpenAI's ChatGPT-5 Pro, signaling a paradigm shift in AI-assisted research despite ongoing limitations.

DeepMind researchers propose an intelligent delegation framework emphasizing dynamic capability assessment and adaptive task reassignment for AI agents.

Google enhances Gemini 3 Deep Think for complex scientific tasks and launches Aletheia, an AI agent that solved 91.9% of advanced mathematics problems.

Google upgrades Gemini 3 Deep Think for science and engineering. Achieves 84.6% on ARC-AGI-2 and Elo 3455 on competitive coding.

Gemini Deep Think achieves breakthrough performance in solving PhD-level math problems and enabling autonomous research in multiple fields.

Google unveils DialogLab, an open-source framework for authoring, simulating, and testing multi-party human-AI conversations beyond one-on-one interactions.

DeepMind's Aletheia AI achieves breakthrough by solving 13 notoriously difficult Erdős problems, demonstrating unprecedented AI-human collaboration in advanced mathematical research.

MIT Technology Review publishes an in-depth analysis of METR's controversial time horizon plot, which has been widely misinterpreted by both AI optimists and pessimists. The graph, which shows AI models' improving ability to complete tasks over time, has led some to believe AI utopia or apocalypse is imminent. The article clarifies the true meaning of the data and addresses common misconceptions about AI capability measurements and progress trajectories.

OpenAI faces senior staff departures as the company prioritizes rapid ChatGPT improvements over long-term AI research projects like Sora and DALL-E.

CSET report reveals AI companies using systems to accelerate R&D, examining implications for innovation, safety, and governance.

A major study reveals that while AI like GPT-4 can now outperform the average person in creativity tests, the most imaginative humans still hold a significant edge, highlighting a clear ceiling for current AI capabilities.

DeepMind's Demis Hassabis, Anthropic's Dario Amodei, and AI pioneer Yann LeCun present conflicting views on AGI achievability, with LeCun arguing LLMs alone cannot reach human-level intelligence.

Researchers have developed a new AI method called Riff-Diff that transforms enzyme design, creating highly efficient and stable biocatalysts for industrial and medical applications. The findings were published in the journal Nature.

Humans&, a new AI startup founded by former researchers from Anthropic, xAI, and Google, has raised $480 million in a seed round, achieving a $4.8 billion valuation with backing from Nvidia and Jeff Bezos.