Build1 publisher2 min readPublished
Anthropic moves part of its biology work out of simulation and into a Bay Area wet lab
Life sciences head Eric Kauderer-Abrams confirmed the lab in a Reuters account, describing the split between in-house work and outside partners as ordinary biotech practice, while a spokesperson said the site is not there to discover drugs.
The Engineer · Build desk
What happened
- Anthropic set up a laboratory for physical biological experiments in the San Francisco Bay Area without announcing it, and its life sciences head, Eric Kauderer-Abrams, confirmed the site exists.
- The company's biological work had rested largely on computational modelling until now, and it has said it wants to work on treatments for rare diseases.
- In June, Anthropic said it would launch its own drug development programmes, concentrating on preclinical research in areas that traditional pharmaceutical manufacturers treat as financially unattractive.
- Novartis chief executive Vas Narasimhan joined Anthropic's board, the company acquired the startup Coefficient Bio in stock, and it released science software for Claude.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint Most drug candidates fail clinical trials on safety or efficacy, so a faster bench loop mainly increases how many candidates arrive at the expensive stage; it does not touch the failure rate there.
- exposure The firm that documented its own systems being usable to develop biological weapons now handles physical biology in-house, which puts its internal materials controls in scope for anyone scrutinising the policy.
- decision With capacity of its own, every experiment becomes a make-or-buy call against contract labs, and Anthropic says it will keep outsourcing where that is more efficient.
- contradiction The account prices Coefficient Bio at about 400 million dollars in stock while also reporting that Anthropic confirmed the deal without disclosing its value, so anyone citing the number is citing the reporting.
A physical lab changes what model output gets graded against. Computational work ends with a ranked list of candidates. A bench assay ends with a result that either agrees with the list or does not. Kauderer-Abrams said that in biology the final verification will for a long time still happen in a real lab, and that the company already does this today [6]. That reaches English through mezha.net's Ukrainian rendering of a Reuters report [2], so treat the wording as a translation and the substance as reported [24].
He described the arrangement as typical for most biotech companies: some of the work in its own facilities, some with outside partners [5]. The most concrete indication of scale in the account is a hiring line rather than a facility spec. Anthropic is looking for a leader to take on the scaling of procurement and other operational processes, and its open roles have included a specialist position [22]. Procurement is what sets a lab's schedule. An assay whose consumables have a ten-week lead time does not run this month.
The automation plan is where the model would touch the bench directly. The company is examining robotic systems driven by Claude to run experiments with minimal human intervention, and says human oversight is still required on safety grounds [12]. Kauderer-Abrams said the use of AI to automate lab work is only at an initial stage, and that the direction could significantly accelerate a great many different processes [13].
For chief executive Dario Amodei the area is personal. In a recent essay he wrote that his father died of a disease a few years before a way to treat it appeared [10]. The lab comes as Anthropic prepares to raise more money through a possible initial public offering at a valuation of up to two trillion dollars [16].
The same company has published the uncomfortable half of this. Several Anthropic researchers recently warned that the technology could lead to human extinction [14], and the company has found examples of its own systems being usable to develop biological weapons [15]. A spokesperson said the lab is not intended directly for discovering new drugs, and declined to say more about what it is for [7]. That leaves the remaining set of possibilities fairly large. Kauderer-Abrams said Anthropic is trying at the same time to raise its models' capability for accelerating drug development and to limit the risks of their use [17].
What to watch
- Whether Anthropic ever describes what the lab actually runs, after a spokesperson declined to detail its tasks.
- Whether the procurement leadership role is filled, and what further lab and operations vacancies follow it.
- Whether the Claude-driven robotics move from examination to a running loop, and what human sign-off sits inside it.