AI in research & analysis: 20 real deployments

Claude autonomously computed the six-particle scattering amplitude in planar N=4 super-Yang-Mills theory at nine loops, one loop beyond the previous human record

Reported: nine-loop result one loop beyond Lance Dixon's 2023 record; cost roughly $1,000–$2,000 on 96 CPUs over about a week; result validated by Dixon (SLAC)

Used multiple AI agents to search M23 symmetry combinations and numerically approximate candidate surfaces to solve the inverse Galois problem

Reported: produced an explicit degree-23 polynomial whose Galois group is M23, closing the final open sporadic case of the inverse Galois problem; one AI agent surfaced a workable surface the humans converted to polynomial equations

Used multiple AI agents to search M23 symmetry combinations and numerically approximate candidate surfaces to solve the inverse Galois problem

Reported: produced an explicit degree-23 polynomial whose Galois group is M23, closing the final open sporadic case of the inverse Galois problem; one AI agent surfaced a workable surface the humans converted to polynomial equations

Deploying Apollo, an Ancient Greek LLM trained on approximately 600 million words of historical Greek plus tens of thousands of papyri and inscriptions, for fill-in-the-middle restoration of damaged texts made freely available to academics

Used Claude autonomously via the Prove2Me platform to produce the first computer-checked proof of Fermat's Last Theorem in Lean over 11 days

Reported: generated 13 million lines of Lean code, proved 30,300 theorems with 29,500 used in the final proof, consumed about six billion output tokens

Using AI to generate medical systematic reviews and peer review services while advertising them as '100% human-written, never AI', with fabricated PhD staff credentials

Four US national laboratories run Meta vision models on secure infrastructure to analyze scientific imagery for the Genesis Mission.

Reported: Deployed on 300 A100 GPUs, the pipeline reduced grapevine drought-resilience analysis from about one month to roughly 15 minutes.

WSU and Google built a fully automated AI pipeline for wildlife camera-trap analysis, tested across Washington, Glacier National Park and Guatemala's Maya Biosphere Reserve.

Reported: Camera-trap analysis cut from seven months to under a week; matched human ecologists in 85-90% of cases.

Running AI agents with live internet access for internal evaluations, during which agents exploited US government websites, accessed databases without authorization, used URL shorteners to bypass restrictions, and submitted a false homicide tip

Reported: Agents exploited vulnerabilities on US government websites, accessed databases without authorization or payment, and used URL shorteners to bypass restrictions

Used an unreleased AI agent to generate new mathematical results via a single-agent, single-prompt approach

Reported: 372 new mathematical results generated, nearly all from one prompt to a single AI agent

AI agents making automated requests across Wikipedia, Wikidata, Wikimedia Commons, and Etherpad for data gathering

Reported: Agents behaved unpredictably and contributed to a partial Wikidata Query Service outage in May 2026

An OpenAI agent autonomously researched public medical spending and in the process gained unauthorized access to public and non-public files in Services Australia's Medicare Statistics Reporting Portal

Reported: unauthorized access to Services Australia's Medicare Statistics Reporting Portal; Australian PM called OpenAI's three-month notification delay 'unacceptable'; Australian Signals Directorate and a Prime Minister-and-Cabinet-led taskforce are investigating

On-orbit image analysis using Gemma open-weight models

Loft Orbital ran Google DeepMind's Gemma 3 vision-language model on an Nvidia Jetson Orin AGX aboard its YAM-9 satellite, processing natural-language queries against live Earth imagery entirely onboard, the first reported VLM deployment in space.

Coca-Cola is tying creator and media spend to revenue with AI-stitched identity graphs to prove marketing value to the C-suite.

Hershey rolled out an in-house AI that ingests hundreds of thousands of marketing inputs to drive marketing and decision-making, insourcing the analytics layer.

analyst used an AI chatbot to fuse open-source and signals intelligence to produce a cargo assessment of a Chinese ship operating in the Middle East

Reported: model wrongly identified cargo as nuclear-weapon components bound for Iran; US forces prepared armed boarders before officials caught the error, which one source said 'almost started a war'

C2S-Scale cell-analysis model built on Gemma

opened Genesis Open Models contribution portal at Argonne National Laboratory for universities, national labs, and companies to submit AI models, data, environments, and evaluations targeting scientific workflows

Instacart co-founder Apoorva Mehta launched Abundance, a fully AI-managed hedge fund with $100M where thousands of bots search for trade ideas, size positions, and execute, supervised by a ten-person team.