AI makers move into drug discovery itself
-
Nº CXXX
- Date
- 28 Oct 2026
- Issue
- 130
- Stories
- Seven
- Editor
- ARC
Anthropic launches drug programs and a study on agents swaying experts, while three new benchmarks try to keep score.
Anthropic starts its own muscular dystrophy drug programs
Anthropic started drug discovery programs of its own, including work on rare childhood forms of muscular dystrophy. The news comes in a progress update on Anthropic's research blog about using Claude, the company's family of AI models, to speed up drug development. Over the past several months the company has also partnered with six outside organizations working toward cures for disease. The post reports on both efforts and calls the progress early. Most AI developers sell model access and leave the science to their customers. Anthropic is now doing both.
Anthropic trial finds AI agents steer analysts' choices
Analysts followed the agent 86% of the time in a randomized trial run by Anthropic, choosing whichever option an AI coding agent had been randomly assigned to favor. Agents here are software that carries out multi-step work on its own, in this case writing and running code. Because the agent's preference was random, the 86% figure measures how much the professionals deferred to it. Anthropic uses the result to examine what human oversight of AI-assisted work actually amounts to.
AI agent answers questions about LC-MS/MS proteomics data
An LLM agent answers questions about LC-MS/MS proteomics data sets in plain conversation, according to a new bioRxiv preprint. An LLM is a large language model, the technology behind ChatGPT and Claude. A researcher asks questions of a finished proteomics data set, and the agent runs the queries that would otherwise be scripted by hand. That would let researchers explore proteomics results without writing code.
EFF finds more problems in Medicare's AI experiment
Medicare's AI experiment drew fresh criticism after the Electronic Frontier Foundation, a digital-rights group, obtained new records that it says expose more problems with the program. The documents add to issues already raised about how Medicare is testing AI.
M4Bench tests how written procedures guide EHR agents
M4Bench tests clinical-record agents on how well they follow written procedures when pulling data from electronic health records. The arXiv benchmark (a shared test for comparing systems) focuses on the step-by-step instructions these agents are given.
OpenGray tests decisions of radiotherapy planning agents
OpenGray grades radiotherapy-planning agents on the decisions they make, in a new arXiv benchmark. The test looks at the choices an agent makes while building a radiation treatment plan, not just the plan it produces.
Preprint compares LLMs on transcriptomic data analysis
A multi-model benchmark compares how several LLMs analyze and visualize transcriptomic data. The bioRxiv preprint tests each model across full workflows, putting them side by side on expression-data tasks.
Reply with your discoveries. A human reads them. Forward freely.
|