Covering a fast-moving AI story means keeping track of the primary documents behind it. The AI company has confirmed it is running a physical laboratory in the San Francisco Bay Area where its Claude model directs robotic experiments, with a stated focus on fundamental biology and rare-disease research. Anthropic’s own lab announcement, published September 23, 2026, adds a specific early result, and several outlets reported the lab in mid-September 2026.
What Anthropic confirmed about its lab
In a Reuters interview reported by CNBC on September 18, 2026, Anthropic’s head of life sciences, Eric Kauderer-Abrams, confirmed the facility. He said, "We believe that to do biology, the final test is still and will be for a while in real lab work," and added, "We absolutely are doing that today." He described the setup as typical of a biotech company, with some work done in Anthropic’s own facilities and some with external partners. A company spokesperson later clarified that the lab is "not for drug discovery specifically," and said human oversight and involvement remain essential for safety.
The lab emphasizes automation, with Claude directing robotic units to run experiments with limited human intervention. TechCrunch reported on September 18, 2026 that Anthropic bought the stealth biotech Coefficient Bio in April, said the lab’s main focus is fundamental biology rather than drug discovery, and announced a joint drug-discovery deal with Novo Nordisk. The source article adds that Anthropic has launched science software, added Novartis CEO Vas Narasimhan to its board, and bought a biotech startup for $400 million in stock. CEO Dario Amodei has said he wants the company to target conditions he considers "undruggable" because they are too expensive to treat, and told TechCrunch, "I believe that AI could cure most major diseases in the next 5-10 years."
What did Claude find in the enzyme search?
According to Anthropic’s lab announcement of September 23, 2026, the team prompted Claude to search a large DNA-sequence database for unusual examples of reverse transcriptase, an enzyme that copies RNA into DNA. Anthropic said roughly 950 agents searched the data for 21 hours and used 210 million tokens, and that the company’s involvement was limited to the initial prompt and the lab work. One agent spotted a repeating pattern of DNA sequences next to the gene for an odd-looking reverse transcriptase.
After analysis and testing in the lab, Anthropic said the pattern marked a previously uncharacterized enzyme system in bacteriophages, the viruses that infect bacteria, which the team named array-associated reverse transcriptases, or ART. The underlying reverse transcriptase, found in a jumbo phage, had been identified in earlier studies, Anthropic said, but Claude appears to be the first to notice the system’s defining features: an associated array of non-coding DNA sequences and an accessory protein of unknown function. The company said it does not yet know the system’s function. Feng Zhang, a professor at MIT and the Broad Institute and a pioneer of CRISPR genome editing, reviewed the preprint and said, "This is an exciting example of how AI agents can contribute to biological discovery. The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation."
The protein-binder benchmark
A second, separately dated result comes from an automated-testing provider. In a case study published August 20, 2026, Adaptyv Bio said the Anthropic team used Claude in Claude Science to design protein binders against 16 targets, then sent them for wet-lab testing. Adaptyv Bio reported that out of 1,320 designs, 354 bound their target, an average hit rate of 26.8 percent, and that a single target yielded zero binders. Adaptyv Bio also said the designs would have won five of six of its Protein Design Competitions and produced tighter binders. These are binding measurements.
How have independent observers received the claims?
The ART result drew outside scrutiny over how independently the AI reached it. The Irish Times, on September 28, 2026, examined that question and quoted Anthropic’s statement: "We are not aware of any previously published work describing the ART system we recently found. Claude was also not trained on any user transcripts, and our molecular biology team has no such access, either."
On the broader capability trend, the UK AI Security Institute’s Frontier AI Trends Report, drawing on two years of frontier-model testing, found that in chemistry and biology AI models first reached its expert baseline for open-ended questions in 2024 and now exceed it by up to 60 percent. The report also found that models able to generate experiment protocols judged accurate first appeared in late 2024, have since been proven feasible to implement in a wet lab, and that today’s systems are up to 90 percent better than human experts at providing troubleshooting support for wet-lab experiments. Those are measured trends across many systems, not a claim about any single discovery.
Why physical experiments still matter
Software can compare sequences and suggest relationships, but a laboratory is needed to confirm that a proposed system exists and behaves as predicted. As Technology Networks noted on September 30, 2026, candidate molecules identified by virtual screening still require experimental validation to determine whether the predicted activity is replicated in biological assays. The publication traced in-silico tools back to a 1977 Nature paper by J. Andrew McCammon and Martin Karplus that reported the first molecular dynamics simulation of a protein. The enzyme project reflects that same loop: Claude searched a dataset and surfaced a candidate pattern, and Anthropic’s scientists handled the physical testing.
What remains unproven
The lab gives Anthropic experimental data, but it does not remove the stages required to turn research into a treatment. Anthropic said the ART system’s function is not yet known and needs further study. CNBC noted that most drugs fail to pass trials for clinical safety and efficacy. The most concrete, verifiable facts for a reader are the scale of the enzyme search, the physical validation step, and Anthropic’s stated focus on fundamental biology.
How to audit a page that covers fast-moving AI news
Rewrite this sentence? The signals worth checking include:
- The sentence names its source in each claim, so a reader can see who reported a figure and when.
- Citations point to primary or high-authority domains, and low-authority or fringe domains do not provide the main support for a factual story.
- Figures carry a date, so you can trace each number back to the report it came from.
- The research is early, not a proven outcome or finished treatment.
- Any AI-assisted draft is reviewed by a person for accuracy before publishing, and a subject-matter editor checks claims that fall outside the site’s usual expertise.
What to do now
- Link each claim on the page to its primary source with a visible date, such as the company announcement, the interview, or the case study, rather than to a secondary rewrite.
- Review the domains you cite and remove any low-authority or fringe sources, replacing them with the original report the claim came from.
- Mark results that are still at an early research stage as such, and avoid headlines or summaries that imply a cure or a product where the source describes a finding.
- Put AI-assisted drafts through a human accuracy check before they go live, and keep a short record of which claims were verified against which source.
- Ask a subject-matter editor to review pages that cover topics outside your team’s core expertise, so a trust problem is caught before it affects the page’s standing in search.
FAQ
What is Anthropic’s wet lab?
It is Anthropic’s first physical laboratory for biological research, confirmed by its head of life sciences in a Reuters interview and located in the San Francisco Bay Area. The company uses its own facilities and external partners, with Claude directing robotic experiments. A spokesperson said the lab is not for drug discovery specifically and that the main focus is fundamental biology.
What did Claude find in the enzyme search?
Anthropic said Claude searched a large DNA database for unusual reverse transcriptases, using roughly 950 agents over 21 hours and 210 million tokens. One agent found a repeating DNA pattern next to the gene for an unusual reverse transcriptase, which the team named array-associated reverse transcriptases, or ART. Anthropic said the system’s function is not yet known.
Is Anthropic’s wet lab for drug discovery?
Anthropic’s spokesperson said the lab is not for drug discovery specifically, and the company said the main focus is fundamental biology. CEO Dario Amodei has expressed his desire for the company to focus on diseases he believes the pharmaceutical industry is neglecting, especially conditions considered "undruggable" because they are too expensive to treat.