Anthropic describes its molecular biology lab in IPO filing and identifies potential liability for workplace injuries and environmental contamination

Anthropic describes its molecular biology lab in IPO filing and identifies potential liability for workplace injuries and environmental contamination

On October 26, Anthropic, the company behind the AI model Claude, submitted a public filing for a share offering on Form S-1 to the U.S. Securities and Exchange Commission (SEC). The company had previously discussed the lab on its website and in interviews with Reuters. It has now included the lab in its disclosures to investors.

In its strategy section, Anthropic says it is building AI agents to help scientists at every stage of drug development. To support this work, it has developed in-house expertise in biology by hiring industry specialists, establishing a molecular biology lab and acquiring Coefficient Bio, a developer of AI models for biological research, in April. The transaction involved 15,4 million Anthropic shares, some of which were compensation for employees who stayed with the company. Media reports put the value at about $400 million in stock.

According to Anthropic’s September 23 post, “human scientists perform all laboratory work” at the lab: Claude generates hypotheses from DNA databases, and people test them experimentally. A separate product connects Claude to equipment: Model Hardware Standard, a protocol available as a research preview. According to the filing, it allows Claude to control microscopes, liquid handlers and commercially available robotic arms. Anthropic considers this approach less suitable for molecular biology, where procedures are designed for each task, although it has tried using it to speed up laboratory work.

In its risk disclosures, Anthropic says that its life sciences program, including its laboratories, and its robotics research extend beyond its previous experience in software development and research. These activities will require it to handle biological and chemical materials, comply with biosafety rules and potentially meet requirements for preclinical research (before human testing) and clinical research.

“These activities expose us to types of liability, including for workplace injuries and environmental contamination, that differ from those associated with our core business.”

Any drug candidates or other discoveries resulting from this work will face lengthy, costly and uncertain development and regulatory pathways and “may never generate revenue.” Connecting models to equipment presents a separate risk: errors, adversarial interference or misuse could cause injuries, equipment damage or the release of hazardous substances, with consequences that “may be immediate and irreversible.” Safety interlocks and human oversight may be insufficient: some components used to connect models to equipment are open source, allowing third parties to use them with untested equipment or remove safeguards. Connecting increasingly capable models to laboratory automation could heighten government and public concerns that AI will accelerate the creation of dangerous biological and chemical materials, prompting further restrictions or scrutiny.

Before the public filing, Anthropic submitted six confidential drafts to the SEC; those drafts are now public. The company’s own lab and all the risks described above first appear in the draft submitted on September 3, two weeks before Anthropic confirmed the lab’s existence in an interview with Reuters. In the September 30 draft, the life sciences program aimed to demonstrate that advanced AI could substantially shorten the time it takes for new medicines to reach patients. That objective had been removed by the October 9 draft and was absent from the public filing.