Anthropic warns of efforts to use its AI to build biological weapons

Free daily briefing on global business news.

The AI company said it could not always determine whether researchers were pursuing legitimate science or weapons development

NurPhoto / Getty Images

Anthropic published a report on Thursday describing efforts by state-linked scientists to use its Claude AI models for research that could support the development of biological weapons, alongside a range of other attempted misuses including cyberattacks, influence operations, and conventional weapons development.

The report covers activity Anthropic detected and disrupted between December 2025 and August 2026. The company said it could not always determine whether the biological research it encountered was legitimate or malicious, because the same techniques that can lead to vaccine breakthroughs can also be used to engineer dangerous pathogens.

"You are not seeing someone in a comic book kind of way say, 'Hey, I want to build a biological weapon to kill everybody,'" Jacob Klein, the head of threat intelligence at Anthropic, told The New York Times. "It's an incredibly nuanced situation," Klein said, according to The New York Times.

Anthropic described five case studies of potentially dangerous biological research. In one example from May 2026, a scientist sought Claude's help writing a grant application for gain-of-function research on chikungunya virus — a mosquito-borne pathogen that causes debilitating symptoms and has no licensed treatment. The proposed research involved engineering mutations to make the virus more harmful as it repeatedly infected live animals. Anthropic said the work was intended to be performed at a military research institute, which heightened its concern.

The company said its newer models have become substantially better at handling sophisticated scientific work than their predecessors, and that this improvement led Anthropic to tighten controls that limit access to queries spanning a broad spectrum of dual-use biological research.

In another case, Anthropic found a researcher outside the U.S. using Claude to plan experiments on highly pathogenic avian influenza focused on mammalian adaptation. Because its biological safety classifiers blocked more capable models from engaging with the material, the researcher's work was confined to Claude's weakest model class, limiting the assistance the system could provide, the company said.

Beyond biological threats, Anthropic said the report documents other categories of misuse. A suspected Russian state-linked group, consistent with public reporting on the actor known as Midnight Blizzard, used Claude to automate cyberattacks against Ukrainian government targets, drone manufacturers, and European diplomatic organizations. The company also identified Chinese and Iranian government-aligned actors using Claude to surveil diaspora communities and dissidents.

Anthropic also flagged what it characterized as an emerging form of misuse: people attempting to enlist Claude in the development of software for conventional armaments such as missiles, armed drones, and bombs. Among the documented incidents, the report identifies three originating in China, two in Russia, and one involving a cell in northern Yemen that test-fired a guided rocket and returned to Claude within hours to analyze why it failed.

The company said it banned accounts associated with all of the activity it identified, shared intelligence with authorities and industry partners, and incorporated its findings into updated safeguards.

Join 500,000+ readers who start their day with Quartz.

By subscribing, you agree to our Terms of Service and Privacy Policy.

Global business news for a smarter world