Join

BRFDaily briefing

Sunday 25 October 2026

The most important AI and AI-for-science stories, read from 26 sources each morning. The same story from several outlets is merged into one item, every item is classified by discipline and by how far its evidence has got, and the list is ranked.

Updated · refreshed daily, about 05:00 UTC

01The briefing

Five stories that mattered in the last 48 hours

01Announcement

Securing our frontier model training environment

OpenAI explains how we strengthened containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.

OpenAI+1 outlet

Why it is hereCovered by 2 outlets · Models & research · Announced by OpenAI

Also covered by The Verge

  1. 02

    Predicting protein electrostatics with protein language models

    PaperAI for scienceNature Communications

  2. 03

    Pennsylvania will require labels on AI-generated ads

    On Thursday, Pennsylvania Governor Josh Shapiro signed the Artificial Intelligence Deceptive Advertising Act into law, requiring disclosures on AI-generated ads.

    NewsAI at largeThe Verge

  3. 04

    Bringing foundational models to the front lines of TB screening

    AnnouncementAI for scienceGoogle Research

  4. 05

    Computer scientists build innovative tech tools to find experiments, detect citation 'hallucinations'

    New York University computer scientists and students have developed a pair of web-hosted AI systems designed to increase the efficiency and accuracy of the research process—by spotting "hallucinations" in academic…

    NewsAI for scienceTech Xplore

02AI for science

By discipline, from the last seven days

ENGEngineering, Earth and climate3

  1. Beyond tree cover: mapping the world’s forest types

    AnnouncementGoogle DeepMind

  2. AI biosecurity must include agriculture

    AnalysisScience

  3. An AI tool improves weather forecasts up to a month ahead

    NewsNature

CHEMChemistry1

  1. 40 years of patent data reveals an overreliance on hazardous solvents

    NewsChemistry World

MATMaterials2

  1. Efficient equivariant framework for universal charge density prediction

    PaperNature Machine Intelligence

  2. New materials library could help AI design better sodium-ion batteries

    NewsTech Xplore

PHYSPhysics and astronomy3

  1. To detect deepfake videos, researchers turn to the physics of light

    NewsPhysics World

  2. Lagrangian–Eulerian learning of flow field and trajectories with TrajectoryFlowNet

    PaperNature Machine Intelligence

  3. Using AI agents to run quantum simulations

    NewsPhysics World

BIOBiology and medicine4

  1. Guidelines for evaluating structure-based methods for protein–ligand binding prediction

    AnalysisNature Machine Intelligence

  2. The Virtual Biotech: A multi-agent AI framework for therapeutic discovery and development

    PaperScience

  3. Sequential diagnosis with language models

    PaperScience

  4. Putting conversational AI to the test in primary care

    NewsNature Medicine

XCross-cutting methods4

  1. Can maths survive AI? Fields medallist Terence Tao calls for a ‘Maths 2.0’ world

    NewsNature

  2. Solving a 31-year-old grand challenge in mathematics – the KLS conjecture

    AnnouncementMicrosoft Research

  3. Researchers develop conversational AI for self-driving labs

    NewsTech Xplore

  4. AI Proved the Quasi-Riemann Hypothesis. What Happens Now?

    NewsQuanta Magazine

03AI at large

Models, industry and policy, from the last three days

Models and research4

  1. AI researchers stranded Claude on a virtual desert island. It started to “play.”

    NewsMIT Technology Review+1

  2. Wearable system captures human activity data for robot training

    NewsTech Xplore

  3. Jack Dorsey’s Block made an AI model that talks to you like an awkward new friend

    NewsThe Verge

  4. Q&A: How AI came to model the “messy” human mind

    UniversityMIT News+1

Industry4

  1. Updated Google privacy policy says your Gmail might be used for advertising

    NewsArs Technica+1

  2. Google’s AI search changes upended online publishing

    NewsThe Verge

  3. Meta delays new business terms after publisher pushback on AI scraping

    NewsThe Verge

  4. Rapidly scaling online storage to serve over 1 billion ChatGPT users (Part II)

    AnnouncementOpenAI

Policy and society4

  1. A timeline of developments in AI safety since the attack on Hugging Face

    NewsTech Xplore

  2. UK pledges more funding to guard against AI risks

    NewsTech Xplore

  3. OpenAI and Poland’s NASK partner to advance AI safety and security

    AnnouncementOpenAI

  4. Intel gets waiver after worries fuzzy US ban on foreign-made robots may hit fabs

    NewsArs Technica

04Cambridge

AI-related news from the University and the Accelerate Programme

  1. Cambridge scientists create a $10 chip that could put a medical lab on your wrist

    Cambridge researchers have created a groundbreaking $10 optical sensor that delivers the precision of laboratory equipment in a device small enough for wearable technology.

    NewsScienceDaily

  2. How can we… use AI to better understand the societal impact of heatwaves?

    Climate change is escalating the frequency and intensity of extreme heat events, which pose a significant threat to public health.

    UniversityEngineeringAccelerate Programme

  3. Researchers receive funding to explore ambitious new ways of improving AI security

    At a time when concerns about AI safety are hitting the headlines almost daily, three researchers here are receiving funding from a £50m UK government programme to explore ambitious new ways of improving AI security.

    UniversityCambridge Computer Science

  4. Undergraduates' work is presented at major conference

    Among the presenters at a major ACM conference today (6 October) will be an undergraduate from this Department, sharing research on ways to improve bias and accuracy in new AI tools that are being developed to assess…

    UniversityCambridge Computer Science

05New research

Peer-reviewed papers, and preprints to watch

Preprints have not been peer-reviewed. Read them as claims, not results.

Peer-reviewed papers8

  1. Predicting protein electrostatics with protein language models

    PaperBiologyNature Communications

  2. Efficient equivariant framework for universal charge density prediction

    PaperMaterialsNature Machine Intelligence

  3. The Virtual Biotech: A multi-agent AI framework for therapeutic discovery and development

    PaperBiologyScience

  4. Sequential diagnosis with language models

    PaperBiologyScience

  5. CauTrigger: deciphering biological system state transitions by hierarchical causal decomposition

    PaperBiologyNature Methods

  6. Lagrangian–Eulerian learning of flow field and trajectories with TrajectoryFlowNet

    PaperPhysicsNature Machine Intelligence

  7. Towards understanding and monitorability of reasoning LLMs

    PaperNature Machine Intelligence

  8. Large-scale iontronic skins with neural-hierarchical processing architecture

    PaperNature Machine Intelligence

Preprints from arXiv8

  1. A novel machine learning framework based on proper orthogonal decomposition for turbine wake flow predictions in natural waterways

    PreprintEngineeringarXiv

  2. Benchmarking Recent Single-Cell Synthetic Data Generation Methods on the OneK1K Dataset

    PreprintBiologyarXiv

  3. Simulation-based inference of profile evolution for high-precision pulsar timing

    PreprintPhysicsarXiv

  4. Unconstrained learning of equivariant atomistic properties

    PreprintMaterialsarXiv

  5. GeoAgent: Contract-Constrained LLM Agents for Seismic Data Processing

    PreprintEngineeringarXiv

  6. Conditional Mean Flows for Frequency-Domain Scattered Wavefield Generation

    PreprintEngineeringarXiv

  7. Wavenumber-Aware Deep Learning for Robust Analysis of Raman Spectra

    PreprintChemistryarXiv

  8. RiboBridge: protein-conditioned de novo RNA binder structure design

    PreprintBiologyarXiv

06Method

How this briefing is made

  1. Read

    Early every morning, at about 05:00 UTC, the site reads 26 public feeds: journals, news outlets, research-lab blogs, Cambridge news and arXiv.

  2. Merge

    Stories that several outlets report are merged into one item. Coverage by several independent outlets is the strongest signal of importance.

  3. Classify

    Each item gets a discipline, and a label for how far its evidence has got: a peer-reviewed paper, a preprint, the developer's own announcement, or a news report.

  4. Rank

    Items are ranked by source, coverage and recency, with promotional pieces pushed down. The reasons appear under each lead story.

Paper
Peer-reviewed research paper
Preprint
Not yet peer-reviewed
Announcement
Published by the developer itself
News
Reported by a news outlet
Analysis
Commentary or newsletter
University
From a university news office
Sources read this morning 26 of 26 available
  • Nature Portfolio: machine learning30 items
  • Nature Machine Intelligence8 items
  • Nature Computational Science4 items
  • Science (journal)5 items
  • Science news0 items
  • arXiv (AI for science)100 items
  • Google DeepMind9 items
  • Google Research11 items
  • Microsoft Research10 items
  • OpenAI62 items
  • Hugging Face25 items
  • Quanta Magazine2 items
  • MIT Technology Review4 items
  • Physics World5 items
  • Chemistry World5 items
  • Tech Xplore (Phys.org)30 items
  • ScienceDaily8 items
  • The Verge10 items
  • Ars Technica20 items
  • Import AI4 items
  • MIT News30 items
  • University of Cambridge1 items
  • Cambridge Department of Computer Science and Technology3 items
  • Accelerate Programme for Scientific Discovery2 items
  • GOV.UK: Department for Science, Innovation and Technology2 items
  • UKRI1 items

Summaries are the first lines of each source's own description; follow the link for the full story. Rankings are automatic and may miss things. Tell us what we should read.