While shouting concerns about “human extinction,” Anthropic quietly built a biological laboratory: speeding up the deployment of AI pharmaceuticals

The Zhitong Finance App learned that Anthropic has built a “wet lab” (wet lab, laboratory for physical experiments) in the San Francisco Bay Area to extend AI ambitions to the field of pharmaceutical science. Two people familiar with the matter said that at a time when public fear of AI is spreading, the startup is advancing biological research from “in silico” (in silico) to the front of real experiments.

Eric Kauderer-Abrams (Eric Kauderer-Abrams), the head of life sciences at Anthropic, confirmed the existence of wet labs in an interview this Tuesday (9/15). “We believe that in biological research, the final testing was and will continue to be a real laboratory job for some time to come.” He said that part of the company's work is done at its own facilities and part is working with external partners, which is a typical model for most biotech companies. The spokesperson later clarified that the laboratory was not dedicated to drug discovery.

From “computer simulation” to a real test bench

Building a wet lab is just one small step in a big ambition. According to a person familiar with the matter, Anthropic wants to overcome diseases it believes the pharmaceutical industry is ignoring, and do so before employees and the public lose confidence that “AI is worth a job or even a human life.” But drug development cannot guarantee success — the vast majority of drug candidates are unable to overcome clinical safety and efficacy testing.

For CEO Dario Amodei (Dario Amodei), the matter also had a personal touch. In a recent article, he wrote that a disease took his father's life, and only a few years later, a cure for the disease appeared. Caudler Abrams shared the same sentiment: “The company's mission is to develop powerful AI in ways that benefit the world. The biggest opportunities we see are in life sciences, and that drives everything we do.”

From selling tools to self-depreciation

Anthropic's intention to operate its own drug pipeline was made public as early as June. At an event in San Francisco, Caudler Abrams said on stage that Anthropic would do pre-clinical work in areas that traditional companies find financially unattractive.

Around this goal, the company has made a series of groundwork: launching a research bench called Claude Science, bringing Novartis CEO Vas Narasimhan (Vas Narasimhan) to the board of directors, and buying startup company Promised Bio with approximately $400 million in shares. Anthropic confirmed the acquisition but did not comment on the deal price.

Today, the company is augmenting its own laboratory capacity to gain first-hand experience in speed and biology, while continuing to outsource more efficient processes. “There are some things we can do on our own much faster.” Caudler Abrams said the goal was to “operate on as large a scale as possible.” The recruitment campaign confirms this: one LinkedIn (LinkedIn) post claims that the company is looking for heads of procurement and operations, one that is recruiting biochemical experts in “protein and nucleic acid characterization,” and another that “our goal is to accelerate progress in life science by an order of magnitude.” Life sciences are already one of Anthropic's biggest investments, he said.

Living with fear of the “extinction of humanity” disaster

Like most biotech companies, Anthropic is embracing physical automation. According to people familiar with the matter mentioned above, the company hopes to advance Claude's ability to command robot units and perform scientific experiments with limited human intervention. However, the spokesperson said Anthropic believes human supervision and participation are critical to safety.

“We're still in the early stages of automating lab work with AI.” Caudler Abrams said, “And this field has the potential to substantially accelerate many different processes.”

This race comes at a time when the company is in a very unusual situation. In the past two weeks, some of the company's researchers warned that AI could lead to human extinction; Anthropic also claimed to have discovered cases where its own systems might be used to develop biological weapons, which raised many questions. Amoudi, on the other hand, wrote on September 12 calling for a slowdown in development, saying that a future AI stronger than the OpenAI agent that broke Hugging Face in July of this year could cause catastrophic damage. Meanwhile, Anthropic is preparing for a potential $2 trillion IPO.

“In our life sciences work, we have always balanced these things.” Caudler Abrams said, “Make the model better in the direction we think will enhance and accelerate drug development, and deploy it carefully and prudently to limit risk.”

This tension is not empty talk. On September 10, Anthropic released a threat intelligence report which stated that five cases of research “may support biological weapon development” were discovered between December 2025 and August 2026. The company also emphasized that “information that can be used to develop biological weapons can also be used to develop vaccines or treatments for certain diseases.”

Aim for “undrugable” targets

His view is that AI can accelerate the development process of diseases previously considered “unruggable” (undruggable), such as speeding up the discovery of bispecific antibodies and tri-specific antibodies — such complex molecules can act on multiple sites on target proteins or cells, and can also hit multiple targets at the same time.

However, it is still unclear exactly which diseases Anthropic is overcoming and how it is progressing. Once the molecule is identified, it will take years of human testing before it can be brought to market, a challenge that Anthropic has yet to take up. Competitor Isomorphic Labs, a subsidiary of Alphabet, has invested in this for many years and postponed clinical targets from the end of 2025 to the end of 2026 — according to the tech media ai2.work, the company has raised a total of about 2.7 billion US dollars, but as of mid-2026, not a single patient is taking medication.

Robotics, hardware standards and pharmaceutical customers

In the automation segment, Anthropic launched a research preview of the “Model Hardware Standard” (MHS) on August 27. According to the company's official blog, the standard connects devices such as microscopes, liquid handlers, and robotic arms to AI agents, reducing hardware integration that used to take weeks to months to “hours or minutes.” Early testing methods include GeneTec (connecting liquid handlers, robotic arms, and enzyme markers in protein testing), Carnegie Mellon University (integration time is about 8 hours, experimental operation is about 3 times faster), and quantum computing company QuERa (laser locking autonomous recovery rate of 99.3%); manufacturers such as Amazon Cloud Technology, Danaher, and QIAGEN have joined the support.

Anthropic is delivering such results along with a wider range of AI services to leading pharmaceutical companies, including Roche (Roche) Gentec, BMY.US (BMY.US), and NVO.US (NVO.US). This bond has just deepened: on September 16, Novo Nordisk announced a partnership with Anthropic to support scientific reasoning in research and development using its cutting-edge models and Claude Science. Amoudi said, “AI capabilities are constantly improving, and it is possible to reduce biological and medical breakthroughs that would have taken a century to less than ten years.”

A trust issue that cannot be circumvented

Meanwhile, Anthropic is dealing with a trust issue — said the two people familiar with the matter. Its customers may be concerned that Anthropic will learn from their competing drug pipelines, even though the AI startup keeps customer data out of its sight.

Caudler Abrams said that due to competitive considerations, Anthropic has drawn a boundary: it does not currently conduct clinical trials and focuses on needs that the industry cannot meet. “We don't compete with pharmaceutical and biotech companies that make money by bringing drugs to market.” he said.

racetrack background

This layout falls on a racetrack that is heating up. According to the industry report “AI Pharmaceuticals Entering the Asset Realization Era”, about 60 billion US dollars have been invested in AI drug discovery since 2019, and about 175 original AI research projects have entered human trials, but none have been approved for listing.

Industry-side cashing signals are appearing simultaneously: According to reports, Ren Feng, co-CEO and chief scientific officer of InsiSi Intelligence (03696), Ren Feng, the company developed in early September, completed the first phase III clinical trial patient administration at Peking Union Medical College Hospital and Shanghai Pulmonary Hospital - this is the first drug candidate in the world to complete target discovery and molecular design by AI. For idiopathic pulmonary fibrosis, it is expected that top-line data will be read in 2029; the molecule takes 1.5 years to synthesize 78 compounds with an investment of 260 At $10,000, the traditional method would take 4.5 years, tens of millions to hundreds of millions of dollars. The policy was added on the same day: On September 18, 10 departments including the Ministry of Industry and Information Technology issued the “Fifteenth Five-Year Plan for the Development of the Pharmaceutical Industry”, which included AI pharmaceuticals among the key tasks of the five-year plan for the pharmaceutical industry for the first time.

Anthropic's own capital market trends are also accelerating. According to reports, the company plans to raise up to 100 billion US dollars through the IPO, with a target valuation of about 2 trillion US dollars. Nvidia is negotiating to invest up to 10 billion US dollars as a cornerstone investor; the company secretly submitted S-1 documents and selected NASDAQ in June. Anthropic told investors that second-quarter revenue was over $11.5 billion, with an annualized revenue operating rate of over $65 billion.

Institutional views

Domestic brokerage firms have viewed “tech giants entering the market” as a core variable for AI pharmaceuticals. Dongwu Securities pointed out in the Pharmaceutical Weekly Report on September 14 that AI pharmaceuticals are entering the “strategic entry stage”. The closed loop of dry and wet is the core logic of AI drug discovery (AIDD). Wet experiments are irreplaceable, and upstream gene synthesis, protein expression and purification, in vitro verification, and model animals will systematically benefit from the explosion of AI candidate molecules. BOC Securities, on the other hand, summarizes AIDD as a closed loop of “design-construction-testing-learning”, and believes that experimental verification is an important part of achieving long-term value.

Overseas judgments are much more careful. Frank von Delft (Frank von Delft), professor of structural biology at Oxford University, suggests that AI models “are not close to making experiments unnecessary” — drug candidates still require expensive real-world tests to verify efficacy and toxicity. University College London professor Matthew Todd (Matthew Todd) determined that any return from Anthropic's own pipeline would take at least ten years.