A source-backed timeline of observable evidence on whether AI is accelerating scientific progress.
Latest report · 2026-08
AI-assisted mathematics produced inspectable proofs, while research agents reached laboratory execution and increasingly demanding biological analysis tasks.
Limitation or failure · Notable · Materials science · Chemistry · Computer science
A replicated open-ended materials-agent study separating stable frontier recovery from a shared database-driven false champion.
Capability benchmark · Notable · Chemistry · Medicine · Materials science
A peer-reviewed retrosynthesis model evaluated on rare reactions, distribution shifts and blind expert ratings of multistep routes.
Measured acceleration · Notable · Biology · Computer science
Optimized inference implementations for over 30 protein and genomics models, with runtime and memory evaluations.
Field diffusion · Notable · General science · Economics
Task-level measurements of scientific AI adoption, complementary model uses and reported workflow bottlenecks.
Capability benchmark · Notable · Biology · Computer science
A peer-reviewed evaluation of agents that expose paper code and data as tested executable tools for reuse.
Direct discovery · Notable · Materials science · Chemistry
LiCoO2 films and an experiment-linked growth map identifying a 650–690 °C high-crystallinity window.
Infrastructure · Milestone · Biology · Medicine
A genome-wide catalogue of predicted molecular effects for nine billion single-nucleotide variants, accompanied by variant-prioritization and experimental research case studies.
Landmark discovery claim · Milestone · Mathematics · Physics
A proposed smooth, compactly supported divergence-free initial velocity on R³ whose unforced incompressible Euler solution loses regularity in finite time.
Direct discovery · Notable · Mathematics · Physics
A two-dimensional inviscid Boussinesq construction with smooth forcing, bounded temperature, and finite-time divergence of temperature gradients and vorticity.
Field diffusion · Notable · Computer science
Internal telemetry reports increased coding-agent use, code contribution and experiment activity across an AI research organization during 2026.
Measured acceleration · Milestone · Mathematics · Computer science
A complete Lean proof of Fermat’s Last Theorem with standard axioms and a statement checked against Mathlib.
Infrastructure · Milestone · Biology · Medicine
A peer-reviewed, openly accessible male Drosophila central-nervous-system wiring diagram with cell typing and sex-difference analyses.
Direct discovery · Milestone · Mathematics
A stronger asymptotic lower bound for the largest consecutive-prime gap below X, accompanied by a public Lean formalization.
Infrastructure · Notable · Climate · Computer science
A peer-reviewed methane detection model plus a released global plume database, trained model, synthetic dataset and inference library.
Direct discovery · Notable · Biology · Medicine
An immune-checkpoint signal in invasive lobular breast cancer, supported by an independent cohort and multiplex immunofluorescence.
Direct discovery · Milestone · Mathematics
A claimed bound liminf(p[n+1] − p[n]) ≤ 186, with a paper, conditional Lean development, and numerical certificate.
Capability benchmark · Notable · Materials science · Chemistry · Computer science
XRDBench combines 100 diagnostic tasks with 34 executable diffraction-analysis workflows.
Measured acceleration · Milestone · Materials science · Biology · Medicine · Computer science
Execution-grounded studies spanning first-attempt CVD growth of three monolayer semiconductors, wet-lab-checked phenotype prediction, and autonomous agent-architecture search.
Capability benchmark · Notable · Biology · Computer science
A 20-study benchmark with 67 GB average raw inputs and 138 graded artifacts for end-to-end computational-biology analysis.
Capability benchmark · Notable · Climate · Medicine · Computer science
An end-to-end geospatial system evaluated on health, risk, food-security, and outbreak-prediction tasks.
Direct discovery · Milestone · Mathematics · Computer science
A new unconditional lower bound showing that more than two thirds of the nontrivial zeros of the Riemann zeta function lie on the critical line, with a public Lean certificate.
Measured acceleration · Milestone · Climate · Physics
A global operational ensemble model for tropical-cyclone track, intensity, and wind-structure forecasting, evaluated on storms from 2023 through 2025.
Landmark discovery claim · Milestone · Mathematics · Computer science · Physics
Ten claimed advances that resolve or substantially advance longstanding problems across mathematics and theoretical computer science, each accompanied by a Lean certificate.
Limitation or failure · Notable · Computer science · General science
Two expert-graded shadow evaluations of frontier agents pursuing the central research questions of unpublished NeurIPS submissions.
Direct discovery · Notable · Computer science · Mathematics
New cryptanalytic attacks that reduce HAWK's effective key strength and improve attacks on a seven-round AES variant.
Field diffusion · Notable · General science · Social science
A global survey of 10,480 researchers measuring AI use, access, satisfaction, and trust across 117 countries or economies and 12 disciplines.
Direct discovery · Notable · Mathematics · Computer science
Five new results in Banach space theory, including proofs for selected open problems, with arguments subsequently checked and refined by the authors.
Direct discovery · Notable · Materials science · Chemistry · Physics
Six CaMg₂Si₂ crystal candidates that passed surrogate stability screening, with three polymorphs receiving post hoc lattice-dynamics analysis.
Capability benchmark · Notable · Biology · Medicine · Computer science
A peer-reviewed benchmark pairing 15 expert-labeled single-cell datasets with 193 data-driven discovery questions derived from 41 published single-cell studies.
Capability benchmark · Supporting · Computer science · General science
An interactive benchmark for agent experiment design under selection bias, measurement error, and hidden confounders.
Capability benchmark · Supporting · Chemistry · Materials science · General science
A 200-question benchmark for agents operating deterministic aqueous-geochemistry simulations.
Direct discovery · Notable · Biology · Chemistry · Medicine
De novo proteins that bind exatecan and apixaban, with experimentally measured nanomolar-to-picomolar affinity and an exatecan binder that protects the drug from hydrolysis.
Capability benchmark · Notable · Biology · Medicine
An expert-authored benchmark of realistic life-science research tasks spanning evidence, analysis, design, validation, translation, and communication workflows.
Capability benchmark · Notable · Physics · Chemistry · Computer science
A 200-question physical-science benchmark spanning physics and chemistry tasks such as reasoning, extraction, experimental design, and code generation.
Limitation or failure · Supporting · Biology · Chemistry · Physics · Mathematics · Computer science · General science
A science-specific safety benchmark covering risk dimensions across multiple scientific disciplines
Direct discovery · Supporting · Mathematics
Proofs of H1 and H2 over prime fields and H3 over arbitrary finite fields, with conditional Lean formalizations of their components.
Capability benchmark · Notable · Biology · Medicine · Computer science
An interactive benchmark of approximately 200 scientific-agent tasks across drug discovery, omics, EHR modeling, and genetics.
Limitation or failure · Notable · Biology · Computer science
A diagnostic benchmark of AI agents over paper-derived scientific inquiry trees
Direct discovery · Notable · Mathematics
An explicit four-map bijection answering the cubic Andrews–Dhar partition question, with accompanying Lean artifacts.
Field diffusion · Supporting · General science · Social science
A publication-level analysis of AI adoption and scientific creativity measures across more than one million papers.
Direct discovery · Supporting · Medicine · Biology · Chemistry
AI-generated compounds that selectively affected target cell types in experimental phenotypic assays.
Limitation or failure · Notable · Computer science · General science
A comparative analysis of AI-generated and human research ideas across scientific fields and historical research landscapes.
Capability benchmark · Notable · Biology · Medicine
A long-horizon spatial biology benchmark requiring agents to recover biological claims from raw or near-raw measurement data.
Direct discovery · Milestone · Mathematics · Computer science
Self-contained proofs and counterexamples for eight questions from published open-problem lists in commutative algebra and Boij-Söderberg theory.
Landmark discovery claim · Milestone · Mathematics
An infinite family of planar point configurations that disproves Erdős's longstanding conjectured near-linear upper bound for the number of unit-distance pairs.
Direct discovery · Notable · Biology · Medicine · Chemistry
De novo protein families that bind cortisol, immunosuppressants, and other small molecules, with structural and experimental testing.
Capability benchmark · Notable · Chemistry · Medicine
A benchmark of guaranteed-solvable, long-horizon small-molecule drug-design tasks across protein targets and chemical task types.
Capability benchmark · Supporting · Biology · Medicine · Computer science
A biomedical machine-learning benchmark with executable end-to-end modelling tasks across heterogeneous data domains.
Adoption context · Notable · General science · Computer science · Physics
NSF X-Labs funds milestone-driven research organizations: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Infrastructure · Notable · Biology · Medicine · Chemistry
Benchling Model Hub embeds scientific AI models in R&D records: capability signal for AI systems on research-adjacent tasks.
Measured acceleration · Notable · Materials science · Chemistry
A closed-loop platform optimized brighter lead-free Cs3Bi2Br9 nanoplatelets with quantified experiment counts and campaign time.
Measured acceleration · Supporting · Biology · Climate
Automated camera-trap image identification that produced multi-species occupancy models similar to expert-labeled workflows.
Limitation or failure · Notable · General science · Computer science · Social science
An audit of non-existent references across papers in arXiv, bioRxiv, SSRN, and PubMed Central.
Limitation or failure · Supporting · Medicine · General science
A biomedical-literature audit measuring fabricated references across PubMed Central papers from 2023 to early 2026.
Field diffusion · Notable · General science · Social science
A large-scale bibliometric measurement of AI adoption, topic concentration, visibility, geography, and retractions across science.
Field diffusion · Supporting · Computer science · General science
Nature surveys early measurements of AI-generated scientific writing: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Field diffusion · Notable · Biology · Medicine · Economics
A downstream field-impact study of how AlphaFold2 changed structural-biology research outputs and target choice.
Limitation or failure · Notable · Medicine · General science
A content-analysis evaluation of AI tools handling retracted scientific articles during literature-search tasks.
Adoption context · Notable · Biology · Medicine
A five-year open-data and technology initiative for predictive models of cells and AI-powered biology.
Capability benchmark · Supporting · Computer science · General science
A benchmark for deep and wide scientific literature discovery tasks with open-ended search requirements.
Capability benchmark · Supporting · Chemistry
Synthegy uses language models for strategy-aware synthesis planning: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Notable · Biology · Medicine
A benchmark for realistic multi-stage scientific data analysis in genomics and quantitative biology.
Measured acceleration · Notable · Biology · Medicine
Medra Lab 001 scales an autonomous biology lab platform: measured or workflow-level change in a research process.
Capability benchmark · Supporting · Biology · Medicine · Chemistry · Computer science
GPT-Rosalind targets life-sciences research workflows: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Biology · Medicine
Artificial allosteric protein switches and biosensors built from machine-learning-designed ligand-binding domains.
Direct discovery · Notable · Biology · Medicine · Chemistry
Amazon Bio Discovery links AI biology models to wet-lab validation.
Field diffusion · Notable · General science · Computer science
Recurring measurement of AI publications, scientific AI capabilities, and remaining expert gaps.
Field diffusion · Notable · General science · Economics · Social science
A peer-reviewed bibliometric analysis of AI diffusion and associations with scientific novelty and impact across more than 170 fields.
Adoption context · Notable · Mathematics · Computer science · General science
A DARPA research opportunity on mathematical foundations for communicating AI-agent collectives in scientific discovery workflows
Limitation or failure · Notable · Medicine · Biology · General science
NIH pilots point toward an AI assurance lab for biomedical research: cautionary signal for reliability, quality, governance, or limits of AI in research.
Infrastructure · Notable · Biology · Computer science
LLM-agent generation and execution of single-cell RNA-seq analyses in notebooks.
Adoption context · Notable · Biology · Materials science · Physics · Chemistry · General science
A funding opportunity to apply AI models and frameworks to national science and technology challenges under the Genesis Mission.
Capability benchmark · Supporting · Materials science · Physics · Computer science
AI agents are tested for operating advanced scientific instruments: capability signal for AI systems on research-adjacent tasks.
Infrastructure · Notable · Climate · General science
CHMv2, an open one-metre-resolution global canopy-height map and accompanying DINOv3-based depth-estimation model for forest monitoring.
Direct discovery · Supporting · Physics · Mathematics · Computer science
Claimed exact analytical methods for gravitational-radiation power spectra from cosmic-string loop geometries.
Capability benchmark · Supporting · Biology
Evo 2 scales genomic foundation models across domains of life: capability signal for AI systems on research-adjacent tasks.
Measured acceleration · Notable · Biology · Medicine · Computer science
DataJoint adds an agentic control layer for reproducible research workflows: measured or workflow-level change in a research process.
Limitation or failure · Notable · General science · Social science · Computer science
A cross-journal analysis measures whether AI-writing policies coincide with disclosure or reduced use of AI-assisted academic writing.
Direct discovery · Notable · Physics · Computer science
Language-model meta-design generates families of quantum experiments.
Adoption context · Notable · General science · Computer science
UKRI sets its first AI strategy for research and innovation: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Limitation or failure · Supporting · General science · Climate
USGS publishes an agency strategy for AI-enabled science: cautionary signal for reliability, quality, governance, or limits of AI in research.
Measured acceleration · Supporting · Medicine · Chemistry
An automated powder-to-tablet workflow generated pharmaceutical formulation data with quantified runtime and data scale.
Field diffusion · Supporting · Biology · Medicine · Economics
A difference-in-differences study of AlphaFold2 effects on structural biologists productivity, citations, journal placement, and inequality.
Adoption context · Notable · Chemistry · Materials science
Review sets criteria for atomistic foundation models in chemistry and materials: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Landmark discovery claim · Milestone · Mathematics · Computer science
An autonomously generated arithmetic-geometry paper, human-AI research papers, and autonomous solutions or partial solutions across a large collection of open Erdős problems.
Capability benchmark · Supporting · Computer science · Mathematics · Biology · General science
A 20-task benchmark for AI research agents using tasks sourced from state-of-the-art machine-learning papers.
Measured acceleration · Notable · Biology · Medicine
Closed-loop autonomous optimization of cell-free protein synthesis reaction compositions with reported cost reduction.
Capability benchmark · Supporting · Biology · Computer science
A more realistic biology research benchmark with nearly 1,900 tasks across practical scientific work contexts.
Capability benchmark · Supporting · Computer science · General science
OpenScholar builds an open retrieval-augmented system for literature synthesis: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Computer science
A benchmark where agents rediscover verified empirical findings from recent machine-learning research papers.
Capability benchmark · Supporting · Physics · Chemistry · Biology
A science benchmark combining olympiad-level problems and PhD-level open-ended research subtasks.
Capability benchmark · Supporting · Biology · Medicine
AlphaGenome predicts regulatory variant effects across genomic modalities: capability signal for AI systems on research-adjacent tasks.
Limitation or failure · Supporting · Computer science · General science
A preprint analysis of hallucinated references in ACL, NAACL, and EMNLP papers from 2024 and 2025.
Adoption context · Notable · Biology · Medicine · Climate · Materials science · General science
A philanthropic AI-for-science funding cohort spanning health, agriculture, biodiversity, fusion data, and carbon-capture materials.
Capability benchmark · Supporting · Computer science · Biology
Deep Research evaluates interactive multi-agent scientific workflows: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Biology · Chemistry · Medicine
Previously unrecognized mammalian metabolites and metabolite families prioritized by a chemical language model and checked against mass-spectrometry data
Field diffusion · Notable · General science · Social science · Computer science
A science-of-science measurement of AI-augmented research across 41.3 million natural-science papers.
Capability benchmark · Supporting · Biology · Computer science
A benchmark framework and single-cell biology instance for open-ended, dataset-grounded scientific agent questions.
Field diffusion · Notable · Biology · Medicine
A quantitative impact study of AlphaFold 2's links to publications, experimental structure submissions, disease-related work, clinical articles, and patent citations.
Limitation or failure · Supporting · General science · Materials science · Physics · Climate
National Academies report recommends hybrid foundation models for DOE science: cautionary signal for reliability, quality, governance, or limits of AI in research.
Measured acceleration · Notable · Physics · Materials science
A self-driving nanophotonics workflow that learned governing equations for steering spontaneous emission from reconfigurable metasurfaces
Field diffusion · Supporting · General science · Social science
A large survey of Max Planck Society and Fraunhofer researchers measuring AI use, tasks, barriers, and perceived research impact.
Adoption context · Milestone · General science · Biology · Materials science · Physics
Federal mission to build an integrated AI platform for national science and technology challenges.
Field diffusion · Notable · Computer science · General science
A large-scale bibliometric measurement of foundation-model use in scientific publications.
Adoption context · Milestone · General science · Biology · Medicine · Materials science · Physics
A national strategy sets AI-for-science missions, data, compute, autonomous-lab, and talent actions across five UK priority research areas.
Direct discovery · Notable · Biology · Computer science
Experimentally active de novo anti-CRISPR proteins and toxin–antitoxin systems generated from genomic context, including sequences lacking significant similarity to known proteins.
Direct discovery · Notable · Biology · Computer science
A peer-reviewed bacterial-evolution mechanism for cf-PICI interspecies spread that matched the AI co-scientist top-ranked hypothesis
Capability benchmark · Milestone · Mathematics · Computer science
Lean-verified formal proofs found for hard competition mathematics problems.
Field diffusion · Supporting · General science
A survey measurement of researcher AI adoption, governance confidence, training adequacy, and research-time pressure.
Infrastructure · Notable · General science · Computer science · Biology · Materials science
Long-horizon autonomous discovery cycles with structured world models and traceable reports.
Measured acceleration · Supporting · Medicine · Chemistry
A robot-assisted active-learning workflow that identified lead aceclofenac nanoformulations from an estimated 17-billion-formulation design space.
Adoption context · Notable · General science
European Commission launches an AI in Science Strategy centered on RAISE: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Field diffusion · Supporting · General science
A global survey measurement of researcher AI use, perceived efficiency gains, tool types, and concern levels.
Measured acceleration · Notable · Chemistry
Robots map chemical reaction hyperspaces at scale: measured or workflow-level change in a research process.
Direct discovery · Notable · Chemistry · Materials science
An experimentally tested eight-element formate-oxidation catalyst with improved cost-specific performance.
Direct discovery · Notable · Biology · Medicine · Computer science
AI-generated bacteriophage genomes that were synthesized and shown to infect bacteria in laboratory tests.
Limitation or failure · Supporting · Computer science · General science
Controlled tests of automated AI Scientist workflows for hidden flaws in generated research outputs.
Measured acceleration · Supporting · Climate · Physics · Computer science
A coupled deep-learning Earth-system model that can run long current-climate simulations with much lower compute than conventional models.
Field diffusion · Supporting · Biology · Computer science · Social science
A bibliometric and causal-inference study of interdisciplinary collaboration after AlphaFold adoption.
Direct discovery · Notable · Physics
Unexpected non-reciprocal force behavior inferred from laboratory dusty-plasma particle motion.
Direct discovery · Notable · Biology · Medicine
AI-generated Cas9-like genome editors, including OpenCRISPR-1, experimentally tested in human cells.
Adoption context · Supporting · Materials science · Chemistry · Physics · Computer science
A Cornell-led AI Materials Institute and broader 2025 NSF AI Institutes cohort with materials discovery as a named science focus.
Direct discovery · Notable · Biology · Medicine · Computer science
Experimentally characterized SARS-CoV-2 nanobodies, including two with improved binding to recent JN.1 or KP.3 variants while retaining ancestral-spike binding.
Measured acceleration · Notable · Biology · Chemistry
A generalized autonomous protein-engineering workflow that improved halide methyltransferase and phytase activity through iterative biofoundry experiments
Direct discovery · Notable · Chemistry · Medicine
Experimentally confirmed de novo adenosine A2A receptor ligands, including three nanomolar ligands and two novel chemotypes characterized by co-crystal structures.
Capability benchmark · Supporting · Chemistry
Chemma adapts large language models for organic synthesis tasks: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Materials science
Standardized evaluation for machine-learning energy models used in crystal-stability screening.
Measured acceleration · Notable · Computer science · Mathematics
AlphaEvolve generalizes coding agents for algorithm discovery and optimization: measured or workflow-level change in a research process.
Direct discovery · Milestone · Medicine · Biology · Chemistry
An AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis tested in a randomized phase 2a trial.
Limitation or failure · Supporting · Computer science · General science
A 2025 position analysis of implementation and verification gaps in AI Scientist systems
Capability benchmark · Notable · Biology · Physics · General science
A multi-domain environment and benchmark for visually rich scientific workflows using professional software.
Landmark discovery claim · Notable · Medicine · Biology · Computer science
AI-agent-proposed macular-degeneration hypothesis followed by lab validation and RNA-seq analysis.
Capability benchmark · Supporting · Chemistry · Materials science
OMol25 scales open molecular DFT data for chemistry models: capability signal for AI systems on research-adjacent tasks.
Limitation or failure · Notable · Medicine · Biology · General science
A peer-reviewed meta-research analysis of formulaic NHANES-derived biomedical papers and statistical risks.
Capability benchmark · Supporting · Medicine
AMIE evaluates conversational diagnostic AI with simulated patients: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Computer science
PaperBench measures whether agents can replicate AI research papers: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Biology · Computer science
A bioinformatics benchmark with more than 50 real-world scenarios and nearly 300 open-answer research questions.
Capability benchmark · Supporting · Computer science · Biology · Medicine
AI co-scientist frames hypothesis generation as a multi-agent research workflow: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Biology · Chemistry
De novo serine hydrolase enzymes with experimentally measured catalytic activity and crystal structures matching design models.
Direct discovery · Notable · Materials science
A generated inorganic material synthesized and measured near a target property.
Direct discovery · Notable · Materials science · Chemistry
Universal machine-learned potentials show systematic softening.
Capability benchmark · Supporting · Biology · Medicine
Nucleotide Transformer benchmarks foundation models for human genomics: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Computer science
AI-agent and human performance compared on open-ended AI R&D environments.
Measured acceleration · Notable · Physics · Computer science
Neural quantum-error decoder tested on real processor data and realistic simulations.
Capability benchmark · Supporting · Medicine
TrialGPT uses language models for patient-to-trial matching: capability signal for AI systems on research-adjacent tasks.
Measured acceleration · Notable · Chemistry · Materials science
Mobile robots perform exploratory synthetic chemistry workflows: measured or workflow-level change in a research process.
Capability benchmark · Supporting · Materials science
OMat24 releases open inorganic-materials data and pretrained models: capability signal for AI systems on research-adjacent tasks.
Field diffusion · Notable · General science · Social science · Computer science
A peer-reviewed measurement framework for direct AI use and citation-associated benefits across 19 scientific disciplines and 292 fields.
Direct discovery · Notable · Mathematics · Physics · Computer science
Explicit, verifier-checked Lyapunov functions for polynomial and non-polynomial dynamical systems beyond cases reached by the compared algorithmic solvers.
Adoption context · Milestone · Biology · Medicine · General science
A Nobel-level recognition event for computational protein design and AI-enabled protein structure prediction.
Capability benchmark · Supporting · Computer science
Agent performance on machine-learning engineering tasks from Kaggle competitions.
Capability benchmark · Supporting · Computer science · General science
Agent performance on research tasks extracted from peer-reviewed data-driven science papers.
Adoption context · Milestone · General science · Materials science · Biology · Medicine
A National Research Foundation-led S$120 million programme for AI methods, interdisciplinary collaborations, shared data and compute, and scientific applications including materials and biomedicine.
Limitation or failure · Supporting · Medicine
EMA publishes a reflection paper on AI across the medicinal product lifecycle: cautionary signal for reliability, quality, governance, or limits of AI in research.
Infrastructure · Notable · Medicine · Biology
TxGNN frames drug repurposing as zero-shot graph prediction: capability signal for AI systems on research-adjacent tasks.
Adoption context · Notable · Physics
NSF and Simons Foundation launch astronomy-focused AI institutes: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Capability benchmark · Supporting · Computer science
DSBench tests data-science agents on more realistic tasks: capability signal for AI systems on research-adjacent tasks.
Measured acceleration · Notable · Chemistry · Materials science
Closed-loop transfer uses AI experiments to extract chemical knowledge: measured or workflow-level change in a research process.
Infrastructure · Notable · Computer science · General science
Prototype automated machine-learning research loop covering ideas, code, experiments, plots, papers, and review.
Measured acceleration · Notable · Materials science · Chemistry · Medicine
Few-shot AI designs antibacterial HDP-mimicking polymers: measured or workflow-level change in a research process.
Capability benchmark · Supporting · Computer science · General science
Model performance on realistic scientific coding tasks across natural-science subfields.
Adoption context · Notable · General science · Materials science · Physics · Climate
DOE roadmap for scientific AI infrastructure across national labs, datasets, models, and supercomputing.
Capability benchmark · Supporting · Computer science · Biology
Practical biology research-task benchmark covering literature, figures, databases, and sequences.
Adoption context · Notable · Biology · Medicine · General science
OpenAI and Los Alamos launch bioscience lab evaluations for frontier models: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Direct discovery · Notable · Materials science · Chemistry
Solid-electrolyte candidates narrowed from a large computed search and followed by synthesis and conductivity characterization.
Direct discovery · Notable · Medicine · Biology
Knowledge-graph reasoning generates drug-discovery evidence chains.
Capability benchmark · Supporting · Computer science · General science
LLM-agent performance on expert-curated data-driven discovery tasks.
Capability benchmark · Notable · Computer science · General science
A virtual environment with 120 tasks for hypothesis formation, experiment design, analysis, and conclusions.
Limitation or failure · Supporting · General science
Cross-science assessment of AI opportunities and reproducibility, trust, and governance risks.
Capability benchmark · Supporting · Medicine
Prov-GigaPath scales whole-slide foundation models for pathology: capability signal for AI systems on research-adjacent tasks.
Infrastructure · Milestone · Biology · Medicine
Structure predictions for biomolecular complexes involving proteins, nucleic acids, ligands, ions, and modified residues.
Infrastructure · Supporting · Chemistry · Materials science
A chemistry agent architecture connecting an LLM to calculators, databases, and lab interfaces.
Direct discovery · Supporting · Chemistry · Medicine · Biology
De novo compounds designed for dual MEK1 and mTOR activity and tested in biochemical and cell assays.
Capability benchmark · Supporting · Computer science
SWE-agent tests agent-computer interfaces for repository-level coding: capability signal for AI systems on research-adjacent tasks.
Infrastructure · Supporting · Computer science · General science
Traceable LLM-agent workflow from data analysis to manuscript drafting.
Limitation or failure · Supporting · General science · Medicine · Climate · Social science
EU scientific advisers recommend infrastructure for AI uptake in science: cautionary signal for reliability, quality, governance, or limits of AI in research.
Limitation or failure · Supporting · General science · Medicine
Stanford AI Index adds a science and medicine chapter: cautionary signal for reliability, quality, governance, or limits of AI in research.
Field diffusion · Supporting · Physics · Mathematics · Computer science
Nature Reviews Physics surveys neural operators for scientific simulation: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Limitation or failure · Notable · Physics · Mathematics
Perspective argues for rigor-first AI in theoretical physics and mathematics: cautionary signal for reliability, quality, governance, or limits of AI in research.
Adoption context · Notable · General science · Biology · Materials science · Physics
RIKEN launches an AI for Science Platform Division: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Measured acceleration · Notable · Materials science · Chemistry
Robotic active-learning search for redox-flow-battery electrolyte formulations.
Direct discovery · Notable · Chemistry · Medicine
Synthesizable antibiotic candidates generated and empirically tested against bacterial pathogens.
Adoption context · Notable · General science
European Research Area issues guidelines for generative AI in research: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Capability benchmark · Supporting · Medicine
UNI scales self-supervised learning for computational pathology: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Materials science · Chemistry · Climate
All-natural nanocomposite plastic-substitute candidates found through robotic preparation and active learning.
Infrastructure · Notable · Biology · Medicine
RoseTTAFold All-Atom models biomolecular assemblies beyond proteins: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Computer science · General science
Data Interpreter builds an LLM agent for end-to-end data science: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Biology
ModelAngelo automates atomic model building in cryo-EM maps: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Biology
scGPT adapts generative pretrained transformers to single-cell biology: capability signal for AI systems on research-adjacent tasks.
Measured acceleration · Notable · Medicine
ChatGPT Advanced Data Analysis is tested on clinical-study machine learning: measured or workflow-level change in a research process.
Measured acceleration · Notable · Physics
Deep reinforcement learning helps avoid tearing instabilities in fusion plasma: measured or workflow-level change in a research process.
Infrastructure · Supporting · Chemistry · Materials science
A proposed metrics framework for autonomy, throughput, parameter-space access, precision, cost, and optimization performance.
Measured acceleration · Notable · Chemistry
Programmable reaction engine integrates synthesis and self-optimization: measured or workflow-level change in a research process.
Adoption context · Notable · General science · Computer science
UKRI funds AI research hubs for science, engineering and real-world data: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Adoption context · Notable · Computer science · General science
Shared U.S. AI research infrastructure for compute, datasets, models, software, training, and support.
Measured acceleration · Notable · Chemistry · Materials science
Dynamic knowledge graphs coordinate distributed self-driving labs: measured or workflow-level change in a research process.
Capability benchmark · Supporting · Medicine
MedSAM adapts segment-anything models to medical images: capability signal for AI systems on research-adjacent tasks.
Limitation or failure · Supporting · Medicine · General science
WHO issues guidance on large multimodal models in health: cautionary signal for reliability, quality, governance, or limits of AI in research.
Capability benchmark · Milestone · Mathematics · Computer science
Olympiad-level geometry proofs produced by a neuro-symbolic theorem prover.
Capability benchmark · Supporting · Computer science
InfiAgent-DABench evaluates agents on executable data-analysis tasks: capability signal for AI systems on research-adjacent tasks.
Adoption context · Supporting · Biology · Medicine · Chemistry
Isomorphic Labs signs AI drug-discovery collaborations with Lilly and Novartis: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Infrastructure · Notable · Chemistry
LLM-guided chemistry planning connected to automated experimental execution.
Direct discovery · Notable · Mathematics · Computer science
New constructions for the cap set problem and new algorithms for studied online bin-packing distributions.
Adoption context · Notable · General science
NSF restricts generative AI use in merit review: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Measured acceleration · Notable · Biology
Deep learning designs cell-type-specific synthetic enhancers: measured or workflow-level change in a research process.
Capability benchmark · Supporting · Computer science · General science
PaperQA applies retrieval-augmented generation to scientific literature questions: capability signal for AI systems on research-adjacent tasks.
Measured acceleration · Milestone · Materials science · Chemistry
Closed-loop autonomous synthesis campaign for inorganic materials.
Direct discovery · Milestone · Materials science
Large-scale candidate generation and computational verification for stable inorganic crystal structures.
Capability benchmark · Supporting · Computer science · Biology · Physics · Chemistry
Expert-written graduate-level science questions for testing advanced scientific reasoning.
Measured acceleration · Notable · Climate · Physics
Fast medium-range global weather forecasts benchmarked against established numerical forecasting systems.
Adoption context · Supporting · General science · Computer science
Argonne announces the Trillion Parameter Consortium for scientific AI models: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Measured acceleration · Notable · Biology · Medicine
High-throughput cell-free production and testing of deep-learning-designed antimicrobial peptide candidates.
Capability benchmark · Supporting · Materials science · Chemistry · Climate
OpenDAC releases a large dataset for direct-air-capture sorbents: capability signal for AI systems on research-adjacent tasks.
Adoption context · Notable · General science · Computer science
UKRI launches the first phase of the UK AI Research Resource: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Capability benchmark · Supporting · Computer science
SWE-bench turns real GitHub issues into a coding-agent benchmark: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Computer science
MLAgentBench evaluates agents on machine-learning experimentation: capability signal for AI systems on research-adjacent tasks.
Field diffusion · Supporting · General science · Computer science
A Nature survey of more than 1,600 researchers on AI use, expected importance, and research-integrity concerns.
Capability benchmark · Supporting · Chemistry · Materials science
AdsorbML accelerates adsorption-energy calculations for catalysis: capability signal for AI systems on research-adjacent tasks.
Infrastructure · Notable · Biology · Medicine
Proteome-wide missense variant pathogenicity predictions for possible human single-amino-acid substitutions.
Adoption context · Supporting · General science · Computer science · Biology · Climate · Materials science
Microsoft launches DeepSpeed4Science for AI-enabled scientific models: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Direct discovery · Notable · Computer science · General science
ChatGPT struggles to attach real evidence to domain answers.
Infrastructure · Supporting · Materials science · Chemistry
CHGNet adds charge-informed dynamics to universal materials potentials: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Medicine
RETFound trains a retinal-image foundation model for disease detection: capability signal for AI systems on research-adjacent tasks.
Limitation or failure · Notable · Computer science · General science · Social science
A peer-reviewed survey and case study of data leakage failures in machine-learning-based scientific claims.
Limitation or failure · Supporting · General science · Social science
UNESCO publishes global guidance on generative AI in education and research: cautionary signal for reliability, quality, governance, or limits of AI in research.
Capability benchmark · Supporting · Computer science · Biology
BioCoder evaluates language models on bioinformatics code generation: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Medicine · Biology
Med-PaLM M prototypes a multimodal generalist biomedical model: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Medicine
Med-PaLM tests large language models on medical knowledge tasks: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Milestone · Biology
Experimentally validated de novo protein designs across monomers, assemblies, scaffolds, and binders.
Capability benchmark · Supporting · Mathematics · Computer science
LeanDojo opens infrastructure for retrieval-augmented theorem proving: capability signal for AI systems on research-adjacent tasks.
Limitation or failure · Supporting · General science
OECD report frames AI in science as a policy and infrastructure challenge: cautionary signal for reliability, quality, governance, or limits of AI in research.
Capability benchmark · Supporting · Medicine
REMEDIS improves data-efficient diagnostic imaging models: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Computer science
Faster low-level sorting routines discovered and included in LLVM libc++ for specific cases.
Capability benchmark · Supporting · Biology · Medicine
Geneformer adapts foundation-model transfer learning to network biology: capability signal for AI systems on research-adjacent tasks.
Limitation or failure · Supporting · Computer science · General science
Reference hallucination treated as a measurable failure mode in LLM-generated research support.
Direct discovery · Notable · Medicine · Biology · Chemistry
A targeted antibiotic lead against Acinetobacter baumannii with mechanistic and mouse-model follow-up.
Measured acceleration · Supporting · Physics · General science
Active-learning controlled physical wind-tunnel experiments that identified turbulence-profile equivalence regions with far fewer configurations
Direct discovery · Notable · Medicine · General science
ChatGPT-generated medical content shows high reference error rates.
Capability benchmark · Supporting · Computer science
HaluEval benchmarks hallucination detection in large language models: capability signal for AI systems on research-adjacent tasks.
Limitation or failure · Supporting · Medicine
FDA opens policy discussion on AI in drug development and manufacturing: cautionary signal for reliability, quality, governance, or limits of AI in research.
Adoption context · Notable · General science · Computer science · Climate · Medicine
NSF expands the National AI Research Institutes network: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Measured acceleration · Notable · Biology · Medicine
Antibody affinity improvements found with small language-model-guided variant screens.
Measured acceleration · Supporting · Biology
Active-learning experiments inferred amino-acid requirements for oral streptococci with far fewer tests than exhaustive screening.
Measured acceleration · Notable · Physics · Materials science · Mathematics
Active learning steers neutron spectroscopy measurements: measured or workflow-level change in a research process.
Capability benchmark · Supporting · Physics · Mathematics · Computer science
AI-Descartes combines data and axioms for scientific law discovery: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Computer science · General science
Semantic Reader explores AI-augmented interfaces for reading papers.
Infrastructure · Notable · Biology
ESMFold predicts metagenomic protein structures from language-model representations: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Computer science · General science
LIDA uses language models to generate data visualizations: capability signal for AI systems on research-adjacent tasks.
Measured acceleration · Notable · Biology
Deep learning supports de novo design of artificial luciferases: measured or workflow-level change in a research process.
Measured acceleration · Notable · Physics
Deep learning filters technosignature candidates in SETI data: measured or workflow-level change in a research process.
Adoption context · Notable · General science · Climate · Medicine
U.S. and EU sign an AI public-good research arrangement: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Direct discovery · Notable · Biology · Medicine
Artificial protein sequences generated by a language model and tested for function.
Infrastructure · Supporting · Chemistry
Transition1x targets reactive chemistry for machine-learned potentials: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Infrastructure · Notable · Materials science · Chemistry
A graph neural interatomic potential used for relaxation, dynamics, property prediction, and large candidate screening.
Capability benchmark · Supporting · Computer science
DS-1000 benchmarks language models on data-science code: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Computer science · General science
Galactica trains a language model on scientific corpora: capability signal for AI systems on research-adjacent tasks.
Adoption context · Notable · General science
Schmidt Futures funds a global AI in Science postdoctoral fellowship: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Capability benchmark · Supporting · Mathematics · Computer science
Draft, Sketch, and Prove uses informal proofs to guide formal provers: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Milestone · Computer science · Mathematics
Provably correct matrix-multiplication algorithms found through reinforcement-learning search.
Capability benchmark · Supporting · Biology · Medicine · Chemistry
DiffDock reframes molecular docking as a diffusion problem: capability signal for AI systems on research-adjacent tasks.
Measured acceleration · Notable · Materials science · Chemistry
Robotic experimentation optimizes non-aqueous battery electrolytes: measured or workflow-level change in a research process.
Direct discovery · Notable · Medicine
CheXzero tests label-free chest X-ray pathology detection.
Measured acceleration · Notable · Biology
ProteinMPNN accelerates structure-conditioned protein sequence design: measured or workflow-level change in a research process.
Adoption context · Notable · Biology · Medicine
NIH launches Bridge2AI for AI-ready biomedical research data: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Infrastructure · Supporting · Chemistry · Materials science
OC22 expands open catalyst data to oxide electrocatalysts: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Measured acceleration · Notable · General science · Materials science · Medicine
Report-level evidence that automated research workflows can shorten cycles and reduce experiment counts.
Measured acceleration · Notable · Physics
Reinforcement-learning plasma-shape control deployed on a research tokamak.
Direct discovery · Notable · Biology
FlyWire scales AI-assisted proofreading for whole-brain connectomics.
Capability benchmark · Supporting · Physics · Chemistry · Materials science
DM21 uses deep learning to improve density-functional simulations: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Mathematics · Computer science
Machine learning guides new conjectures in pure mathematics.
Capability benchmark · Supporting · Biology · Medicine
EVE predicts disease variant effects from evolutionary sequence data: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Biology · Medicine
Enformer improves DNA sequence-to-expression prediction: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Mathematics · Computer science
miniF2F standardizes formal Olympiad-level mathematics benchmarks: capability signal for AI systems on research-adjacent tasks.
Landmark discovery claim · Canonical · Biology
High-accuracy computational protein structure predictions validated through CASP14 and published in Nature.
Measured acceleration · Notable · Physics · Computer science
Near-real-time gravitational-wave parameter inference compared with standard Bayesian workflows.
Limitation or failure · Notable · Physics · Mathematics · Computer science
Physics-informed machine learning becomes a framework for scientific modelling: cautionary signal for reliability, quality, governance, or limits of AI in research.
Adoption context · Supporting · Biology · Medicine · Computer science
Broad launches a Schmidt-funded center for biology and machine learning: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Capability benchmark · Supporting · Medicine · Biology · Chemistry
Therapeutics Data Commons standardizes AI tasks for drug development: capability signal for AI systems on research-adjacent tasks.
Infrastructure · Supporting · Biology
cryoDRGN uses neural networks to model cryo-EM heterogeneity: context signal for AI-enabled research infrastructure, adoption, policy, or field diffusion.
Capability benchmark · Supporting · Chemistry
Bayesian optimization is benchmarked for synthetic reaction tuning: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Mathematics
The Ramanujan Machine generates conjectures about mathematical constants.
Capability benchmark · Supporting · Physics · Computer science
Community benchmark for model-agnostic anomaly detection in simulated collider events.
Capability benchmark · Supporting · Medicine · Biology
nnU-Net automates configuration for biomedical image segmentation: capability signal for AI systems on research-adjacent tasks.
Measured acceleration · Notable · Materials science
Closed-loop active learning at a synchrotron beamline for materials phase mapping and optimization.
Infrastructure · Notable · Physics · Mathematics · Computer science
Neural-operator surrogate models for families of PDE solution maps.
Measured acceleration · Notable · Physics · Computer science
MeshGraphNets extend learned simulation to adaptive mesh systems: measured or workflow-level change in a research process.
Direct discovery · Notable · Physics · Computer science
CAMELS builds a simulation suite for machine-learning cosmology.
Measured acceleration · Notable · Chemistry · Materials science
Autonomous robotic search over photocatalyst mixtures for hydrogen production.
Direct discovery · Notable · Chemistry · Materials science
Cu-Al electrocatalysts for CO2-to-ethylene conversion identified through active machine learning and experiments.
Capability benchmark · Supporting · Computer science · General science
SciFact benchmarks evidence-based verification of scientific claims: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Computer science · General science
SPECTER uses citation-informed transformers for scientific paper embeddings: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Physics · Computer science
Equivariant normalizing flows improve sampling for lattice gauge theory.
Infrastructure · Supporting · Physics · Computer science
Graph Network-based Simulators learn complex physical dynamics: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Medicine · Biology · Chemistry
An AI-screened molecule was identified as an antibiotic candidate and tested experimentally.
Capability benchmark · Supporting · Physics · Computer science
OmniFold applies machine learning to full-phase-space collider unfolding: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Computer science · General science
S2ORC provides machine-readable full text for AI literature tools: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Physics · Chemistry
FermiNet applies deep neural networks to many-electron wavefunctions: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Physics · Mathematics · Computer science
Symbolic regression benchmark results on recovering equations from data.
Infrastructure · Notable · Materials science
Crystal graph neural networks predict materials properties from structure: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Chemistry
Neural-symbolic AI plans organic synthesis routes: capability signal for AI systems on research-adjacent tasks.
Capability benchmark · Supporting · Physics · Computer science
CaloGAN explores generative models for fast particle-detector simulation: capability signal for AI systems on research-adjacent tasks.
Direct discovery · Notable · Physics
Deep learning helps identify new exoplanets in Kepler data.
Infrastructure · Notable · Physics · Mathematics
Neural quantum states apply machine learning to many-body wavefunctions: capability signal for AI systems on research-adjacent tasks.