Anthropic Confirms It Runs a Wet Biology Lab in the Bay Area

Anthropic Confirms It Runs a Wet Biology Lab in the Bay Area · Chronopin

Anthropic confirmed on 18 September 2026 that it runs a wet biology lab in the Bay Area where its AI models can run physical experiments; it declined to say where the lab is or exactly what it works on^[1][2][3]
Reuters broke the story at 3:02 AM PDT and TechCrunch carried Anthropic's confirmation to it the same afternoon, at 4:13 PM PDT^[2][3]
"We believe that to do biology, the final test is still and will be for a while in real lab work... We absolutely are doing that today" - Eric Kauderer-Abrams, Anthropic's head of life sciences, who says the lab runs like most biotech labs, with some research in its own facilities and some with external partners^[2]
The stated focus is fundamental biology, not drug discovery: a spokesperson clarified that the lab "is not for drug discovery specifically", declining to elaborate - though Reuters frames it as part of a ramping AI drug program and reports that Anthropic wants Claude to direct robotic units carrying out experiments with limited human intervention^[1][2][3]
Anthropic counts pharma companies as customers - Roche's Genentech, Bristol Myers Squibb and Novo Nordisk, with a joint drug-discovery deal just announced with Novo Nordisk - and has set a boundary that it is not running clinical trials for now, partly over competition concerns^[2][3]
It follows the purchase of Coefficient Bio, a stealth AI biotech, in April, reported at about $400 million in stock; Anthropic confirmed the acquisition but not the price^[2][3]
The missing piece arrived on 27 August: the Model Hardware Standard research preview, a specification connecting agents to programmable instruments - microscopes, liquid handlers, robotic arms - that exposes each device's operations and safety limits and works with models other than Claude^[1][5]
A day before the lab news, on 17 September, Anthropic reported optimising more than 30 biomolecular models in under four weeks with two staff supervising Claude, for about a 4x average speedup where small numerical differences are allowed and about 1.6x on structure prediction with identical outputs; the code is open-sourced on GitHub^[1][4]
Its 18 August protein-design work reported binders for 14 of 15 tested targets, produced and tested by external evaluators Adaptyv Bio and Twist Bioscience, at about $150 of GPU and token spend against campaigns previously allowed up to $10,000 per target - but none of 90 designs bound the hardest target, maltose-binding protein^[1][4][6]
Anthropic also launched a Life Sciences Verification Program the same week, giving vetted biology researchers access to its most powerful models^[3]
The juxtaposition is the story: Dario Amodei has said AI could cure most major diseases in five to ten years and has repeatedly named bioterrorism among AI's biggest risks, while in the previous two weeks Anthropic researchers warned AI could lead to human extinction, the company said it had found examples of its systems being usable toward biological weapons development, and its alignment lead put the odds of AI exterminating humanity this decade above 10%^[2][3]
How far Claude, the Model Hardware Standard and the lab are actually wired together is not disclosed, and the MHS preview's QuEra pilot showed Claude unable to troubleshoot physical hardware failures and pausing overnight for human confirmation^[1][5]

参考资料 6可信度 93%

第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均；资料每比最新的一份旧 180 天，权重减半。

[1]90%
therundown.ai/news/anthropic-claude-biology-lab
therundown.ai· 发布于 2026年9月22日· 开始 2026年9月18日 ✓· 占评分 17%
The source page carries its own date of 22 September 2026 and is a write-up of earlier reporting, so the event's date comes from the two reports it is built on: Reuters^[2] published the exclusive at 3:02 AM PDT on 18 September 2026 and TechCrunch^[3] the confirmation at 4:13 PM PDT the same day - "Anthropic^[4][5][6] has a wet biology lab in the Bay Area where it can use its AI models to run physical experiments, it has confirmed to TechCrunch". The pin is the day the lab was disclosed: Anthropic gave no date for when it opened and declined to say where it is beyond the Bay Area, so the pin sits on San Francisco, the city of its own headquarters.
[2]95%
EXCLUSIVE: Anthropic quietly sets up biology lab as it ramps AI drug program
reuters.com· 发表于 2026年9月18日· 占评分 17%
The original exclusive the source^[1] is built on, by Jeffrey Dastin and Michael Erman, dated SAN FRANCISCO, Sept 18. It carries the Eric Kauderer-Abrams interview ("We believe that to do biology, the final test is still and will be for a while in real lab work... We absolutely are doing that today"), the two people familiar with the matter who place the wet lab in the San Francisco Bay Area, the spokesperson's clarification that it is not for drug discovery specifically, and detail no rewrite carries - Coefficient Bio at about $400 million in stock, Novartis CEO Vas Narasimhan on the board, and the boundary against running clinical trials.
[3]90%
Anthropic is operating a lab that conducts biology experiments
techcrunch.com· 发表于 2026年9月18日· 占评分 17%
Julie Bort's report the same afternoon, and a confirmation in its own right rather than a rewrite of Reuters^[2]: "Anthropic^[4][5][6] has a wet biology lab in the Bay Area where it can use its AI models to run physical experiments, it has confirmed to TechCrunch." It adds the April purchase of Coefficient Bio, the Novo Nordisk joint drug-discovery deal, the Life Sciences Verification Program launched the same week, and the counterweight the source^[1] keeps - Amodei's call to slow down, the alignment lead's greater-than-10% extinction odds and Chamath Palihapitiya's post about a wet lab in SF.
[4]95%
How Claude is uplifting biomolecular modeling
anthropic.com· 发表于 2026年9月17日· 占评分 17%
Anthropic's^[5][6] research post published the day before the lab was confirmed: more than 30 biomolecular models optimised in under four weeks, ~4x average speedup and ~1.6x on structure prediction with identical outputs.
[5]95%
Previewing the Model Hardware Standard
anthropic.com· 发表于 2026年8月27日· 占评分 16%
Anthropic's^[4][6] own announcement of the interface that lets an agent drive microscopes, liquid handlers and robotic arms - the piece of the stack that makes a physical lab usable by Claude, and the source^[1] of the QuEra pilot's limits.