Anthropic built a wet lab to push AI into drug science, Reuters reports

Anthropic on Wednesday unveiled the Life Sciences Verification Program, a beta initiative that for the first time gives vetted researchers access to its most advanced AI models — including its Mythos line — with relaxed biosafety guardrails tailored for biology work. The announcement came alongside a Reuters report that the company has quietly built a wet laboratory in the San Francisco Bay Area, marking a concrete step from computational AI into physical science.

Opening the Gates for Biology Research

The LSVP, announced September 17, allows verified academic labs, startups, and pharmaceutical companies to apply for access to Anthropic's Mythos 5.1, Opus 5, and Sonnet 5 models under two grant tiers. Standard Use grants cover most research and development workflows and renew annually. High-risk Use grants, designed for dual-use research such as viral vector characterization, remove additional safeguards but apply to a single project and require renewal every six months.unite+1

Anthropic said it has already onboarded dozens of organizations through early access and expects to enroll hundreds within the first week. Xaira, Edison, and Manifold Bio are among the first partners to describe plans for the program. Access will later expand to individual Pro and Max subscribers.unite

The program was built against three specific threat models — account compromise, insider threats, and autonomous agent misuse — and shifts enforcement from real-time blocking to offline behavioral monitoring with a 30-day data retention requirement. Anthropic said its September 2026 threat intelligence report documents increasingly sophisticated biological misuse attempts on its platform, including cases it assessed could support biological weapons development.unite

A Wet Lab and Faster Biology Models

Separately, Reuters Thomson Reuters Corporation ↘ 2.33% reported on Friday that Anthropic has established a physical wet lab, confirmed by the company's head of life sciences, Eric Kauderer-Abrams. "We believe that to do biology, the final test is still and will be for a while in real lab work," he told Reuters. Anthropic is also exploring how Claude can direct robotic units to carry out experiments with limited human intervention.reuters

On the computational side, Anthropic published research showing Claude optimized more than 30 open-source molecular biology models in under four weeks, achieving an average fourfold speedup. In a protein design benchmark, Claude matched the performance of traditional campaigns costing roughly $10,000 per target while using a single Nvidia ↘ 0.25% H200 GPU for 24 hours at a cost of about $150. The optimized code was released on GitHub under an Apache 2.0 license.pasqualepillitteri+1

Navigating Trust and Competition

The moves come as Anthropic prepares for a potentially massive IPO and seeks to demonstrate that its AI capabilities extend beyond code and text generation into high-value scientific domains. The company faces a delicate balancing act: it serves major pharmaceutical clients including Roche's ↗ 4.38% Genentech, Bristol Myers Squibb Bristol-Myers Squibb Company ↘ 1.20%, and Novo Nordisk ↗ 0.12%, while simultaneously building its own drug discovery ambitions.reuters

Kauderer-Abrams said Anthropic has drawn a boundary. "We're not competing with pharma and biotech companies that make their business in bringing drugs to market," he told Reuters, adding that the company is focusing on diseases that the industry would not otherwise address.reuters