Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-07-02 · high
Claude Science is meant to run biology and drug-development research from plain instructions, and Jumper's move points to a fight for AI science talent.
Why now: Anthropic introduced Claude Science at its June 30 AI for Science briefing, with Jumper's reported DeepMind departure surfacing in the same coverage.
Anthropic launched Claude Science on June 30, aiming to automate research work in computational biology and drug development, per MIT Technology Review.
The product is pitched as the research version of Claude Code, running lab work from high-level instructions instead of writing code from a prompt. For biotech teams, handing that work to an agent raises the stakes on trusting what it produces.
Coverage ties the launch to John Jumper, the Nobel laureate who created AlphaFold at Google DeepMind, reportedly leaving DeepMind for Anthropic. It's a marquee hire in a lab talent market already realigning around AI science.
Software agents fail loudly. A broken build or a bad diff shows up in code review before it ships.
A research agent running computational biology work can fail quietly instead. A wrong hypothesis or a bad simulation input can shape a paper before anyone catches it.
The pattern is familiar to builders outside biotech too. Claude Code automated software work, and Claude Science is aimed at the same automation for biology. What's still missing from the coverage is any word on how Anthropic verifies a research agent's output before a lab acts on it.
Each link below shares sources, entities, or timing with this story.
The new beta doesn't add scientific reasoning, it adds plumbing, connecting Claude to genome browsers, protein viewers, and chemistry tools so researchers stop copy-pasting between them.
Per MIT Technology Review, Claude Science aims to do for research what Claude Code did for software: autonomously carry out computational-biology and drug-development work from high-level instructions. Coverage ties it to Nobel laureate and AlphaFold creator John Jumper report...
A researcher found hidden Unicode markers in Claude Code's system prompts flagging traffic from DeepSeek, Zhipu, Baidu, and Alibaba, and Anthropic removed the code without saying so in the changelog.
The internal tool that triages Anthropic's build failures ships as reference code that mines your own incident history before it touches anything.
The mirror syncs nightly from Anthropic's internal review queue, and every plugin in it already cleared automated security scanning before listing.
Repos and secrets stay on your network, but prompts and results still leave for inference, and one user locks each runner.
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.