AI giants move in on drug development
-
Nº CXXVI
- Date
- 22 Oct 2026
- Issue
- 126
- Stories
- Five
- Editor
- ARC
OpenAI buys a drug-development agent startup, and three preprints go after the parts nobody demos.
OpenAI buys an agent startup for drug development
OpenAI acquired Orchestra Bio , a startup building agents that help pharmaceutical companies run drug development, the company told Axios. Orchestra's platform lets scientists hand off work like monitoring ongoing studies and tracking progress, and its team and technology will fold into Rosalind, OpenAI's set of models and tools for life-sciences research. The deal comes as OpenAI and Anthropic both push into science from their respective strongholds in coding and security. What gets automated here is the day-to-day running of a drug program, things like study monitoring and progress tracking, rather than target discovery or molecule design, where most pharma AI spending has gone so far.
Local AI agents mine outbreak reports for CRISPR guides
Agents running on local hardware pull candidate CRISPR guide sequences out of outbreak reports, in a new bioRxiv preprint. The agents read published outbreak documentation and propose guides against the pathogens described, with the models running on the researchers' own machines rather than through a vendor API. Running locally is what sets the work apart, since agent work in biology has mostly assumed a model sitting elsewhere at the other end of an API call.
BatchBench tests whether image models survive batch effects
BatchBench scores image models on how badly they break when the data comes from a different batch or site, using the JUMP Cell Painting collection and a public leaderboard. Batch effects are the usual reason a morphological profiling model looks strong on one plate set and mediocre on the next. Putting a published number on that failure makes it checkable instead of anecdotal.
Agents take over multi-omics variant annotation
Agents handle variant annotation across multi-omics data in hvantk, a framework posted to bioRxiv. It wraps annotation and downstream analysis steps that normally run as separate command-line tools into a single agent-driven workflow.
BRIDGE blocks model updates that only look better
An acceptance gate blocks model updates that only appear to be improvements. BRIDGE checks a candidate update against the version already in use before it replaces it, with a retrospective evaluation on clinical data.
Reply with your discoveries. A human reads them. Forward freely.
|