#Science & Research

What is this?

AI for science means models that speed up discovery itself. They propose hypotheses, prove theorems, predict weather, find new materials and read whole literatures in minutes. Many researchers believe this, not chatbots, is where AI changes the world most.

Key tools & players

DeepMind — AlphaFold, GNoME (materials), GraphCast (weather), AlphaProof (math)
OpenAI — frontier models for research; a program giving 100,000 academics free access (our coverage)
Anthropic — Claude for scientific reasoning and security research
FutureHouse and "AI co-scientist" projects — autonomous research agents

Milestones

2020-2023 — AlphaFold2 opens the era, then GNoME and GraphCast take on materials and weather
2024-2025 — AlphaProof reaches Olympiad silver, then LLMs take IMO gold and co-scientist agents deliver lab-validated hypotheses
2026 — Claude finds encryption weaknesses experts missed, and OpenAI opens frontier models to 100k researchers (our coverage)
Aug-Sep 2026 — AI takes the prime-gap record, FrontierMath's hardest tier and a machine-checked proof of Fermat (our coverage)
Aug 2026 — First complete AI-designed genomes: 16 working bacteriophages, published in Science (our coverage)
Aug-Sep 2026 — DeepMind's WeatherNext goes operational at the US Hurricane Center, then WeatherNext 3 delivers hourly 5 km global forecasts (our coverage)
Sep 2026 — Claude finishes a particle-physics calculation one layer past the record a SLAC physicist held (our coverage)
Sep 2026 — Claude turns 30-year-old Magellan radar into a public elevation map of a third of Venus
Sep 2026 — 25 Fields medallists protest as OpenAI answers with an advisory group and 100 solved problems (our coverage)
Sep 2026 — Claude rewrites 30+ biology models in a month, then finds a new phage enzyme system in Anthropic's own lab (our coverage)

Mini glossary

Co-scientist: an AI agent that proposes and helps test hypotheses
Foundation model: one large model adapted to many scientific tasks
Benchmark: standardized test used to compare model abilities

Stories on this topic

More than a fifth of reviewers used AI after being told not to.
2026-09-29 An AI reran 4,452 economics papers and flagged most of them.
2026-09-29 A wrong AI summary cut what people remembered almost in half.
2026-09-29 OpenAI backs a lab making drugs for viruses that do not exist.
2026-09-28 The best model found 78% of the facts in a patient's file.
2026-09-26 Claude pushed a physics calculation past the record a human set.
2026-09-26 AI tuned a virus to infect bacteria a million times more effectively.
2026-09-25 arXiv raised $17.2 million, partly to handle AI-written papers.
2026-09-24 Claude found an unknown enzyme system in a 21-hour search.
2026-09-24 Mathematicians used AI to close the last gap in a 40-year hunt.
2026-09-23 OpenAI's million-dollar proof leaned on a force nature never makes.
2026-09-23 An AI broke an Enigma message unsolved since 2005.
2026-09-22 A model trained only on ancient Greek is restoring broken papyri.
2026-09-22 OpenAI says one model resolved more than 100 open maths problems.
2026-09-22 Terence Tao is building maths models no AI company controls.
2026-09-20 A web developer used AI agents to settle a 50-year-old conjecture.
2026-09-20 Alibaba opened up a model that spots 146 diseases in CT scans.
2026-09-19 Anthropic and a lab will test 5,000 AI-designed proteins for free.
2026-09-19 Anthropic's Claude spent a month speeding up protein design software.
2026-09-19 The Clay Institute says a $1m maths problem is apparently settled.
2026-09-15 A letter from 771 mathematicians pushed OpenAI out of a contest.
2026-09-13 AI agents broke a 35-year-old challenge built to test encryption.
2026-09-13 25 Fields medallists signed one letter against AI problem-solving.
2026-09-12 GPT-6 Astra solved the last problem in maths' hardest test.
2026-09-11 A Fields medallist opened an institute to do the maths of AI safety.
2026-09-10 A mathematician asks if his ChatGPT chats trained OpenAI's prover.
2026-09-10 A mathematician says OpenAI asked him to erase a co-author.
2026-09-10 Terence Tao warns AI is mining open maths problems dry.
2026-09-10 Singapore will fund four years of AI research for new PhD holders.
2026-09-09 An AI agent tuned a real quantum chip at MIT by itself.
2026-09-09 OpenAI's agent swarm cracked a famous fluid equation in 88 hours.
2026-09-09 DeepMind mapped every possible single-letter change in human DNA.
2026-09-08 A quarter of DeepMind's cheating agent swarm turned whistleblower.
2026-09-08 Caltech is giving math teams $2 million of AI compute for a weekend.
2026-09-07 An AI-designed lung drug made patients' blood look years younger.
2026-09-07 Seven minutes with Gemini beat a fact sheet at easing conspiracies.
2026-09-06 Terence Tao says the AI race on primes contaminated the problem.
2026-09-06 Claude wrote 13 million lines of code to check Fermat's Last Theorem.
2026-09-05 Google's new weather model now refreshes its forecast every hour.
2026-09-04 Anthropic's model mapped a third of Venus from 30-year-old radar.
2026-09-03 One developer beat many chatbots on a reasoning test for 67 cents.
2026-09-02 The best AI agent finished 30% of real lab work
2026-08-29 Claude beat human researchers at closing alignment gaps
2026-08-29 The protocol that let AI touch your files now has a sibling for lab equipment — Anthropic wants a model to drive a microscope the way it drives an app
2026-08-28 MIT built a model that forecasts the flood that has never happened — 300mm of rain on New York, where the record is about 200
2026-08-26 The FDA has cleared 1,357 AI medical devices. Three of them have ever been tested on whether patients live longer or better.
2026-08-26 A question mathematicians have not been able to answer since 1948 was answered on Sunday, in a 108-page file posted to X
2026-08-25 By the end of last year, nine in ten biomedical papers carried the fingerprints of a language model
2026-08-24 A 27-billion-parameter model beat Opus 4.8 and GPT-5.5 at reproducing published papers
2026-08-23 Give a model the reference list of an unpublished paper and ask what the paper says. The best ones get it 15 percent of the time.
2026-08-22 Google's open models have been downloaded a billion times, and outsiders have built 100,000 versions of them
2026-08-22 Machine learning read the shape of sick brain cells and picked out nine already-approved drugs that calmed them down
2026-08-21 The FDA has cleared 1,357 AI medical devices. Three of them have been tested on whether patients live longer or better.
2026-08-20 So many machine-written proofs arrived that mathematicians built a filing office — and one of its two checks is done by a language model
2026-08-20 The reason AI has not cured cancer may be that almost none of its training data comes from human tissue
2026-08-19 A general chatbot drove the specialist protein tools itself — and two outside labs confirmed the molecules stick
2026-08-19 Britain will move aircraft over the Atlantic this winter to stop them drawing the clouds that trap heat
2026-08-18 The argument about a runaway intelligence explosion is a claim about equations. An Oxford philosopher just worked them out.
2026-08-18 Two of the world's most decorated mathematicians looked at what AI actually solves, and landed on the same missing piece
2026-08-17 They built a language model that has never read anything above fifth grade — then found they could not teach it more
2026-08-16 A Fields medallist set out what AI is genuinely good at in maths — and the one thing it has no nose for
2026-08-12 OpenAI offered 100,000 researchers free frontier access. The first round has become a lottery.
2026-08-12 A firm selling ‘100% human-written, never AI’ medical research turned out to be AI all the way down
2026-08-12 Proposals that read like AI wrote them win more NIH money — and ask smaller questions
2026-08-12 Anthropic pointed an unreleased Claude at the Riemann hypothesis. It did not solve it — it nearly doubled the best known bound.
2026-08-11 Two rival AI models were asked the same 25-year-old engineering question. Both came back with a proof — and they were different proofs
2026-08-10 The US government wants your data, your model or your fine-tune — first Genesis deadline is August 14
2026-08-09 An extra day of warning: DeepMind opens its cyclone model, and the Hurricane Center is already using it
2026-08-09 Jeff Dean published his startup's pitch deck — and Alphabet is one of the investors
2026-08-08 Mathematicians say OpenAI's AI-written proofs leaned on their work without crediting it
2026-08-08 AI wrote 16 working viruses from scratch — and the study just cleared peer review in Science
2026-08-07 Why Erdős's puzzles keep falling to AI — and what mathematicians say it means
2026-08-06 Two teams gave the same AI the same open problem. Their proofs landed 3 hours apart.
2026-08-03 Fields Medal winner Jacob Tsimerman is joining OpenAI to work on AI safety
2026-08-02 OpenAI's field report: coding agents sped up real science software up to 60x
2026-08-02 OpenAI previews 'Astra' — agent teams that solved ten open math problems for about $2,000
2026-08-01 35-country survey: 88% of university students use AI — and most say nobody taught them how
2026-07-31 First classroom trial of AI for teachers: time saved, student motivation dented
2026-07-31 OpenAI is giving 100,000 academic researchers free frontier AI access
2026-07-31 Claude found encryption weaknesses that years of expert review missed
2026-07-31 Read in the app — free, in 9 languages