Anthropic says it blocked use of Claude in research with bioweapon risk

Anthropic says it has blocked several attempts to use Claude, its artificial intelligence assistant, in research that could help create biological weapons. The information is in a new company safety report. According to the company, the blocks are part of dozens of interventions against potentially malicious uses since December 2025, including access from outside the United States.

The document describes five case studies in which Anthropic's models were used in a way that, in the company's assessment, could support the development of biological weapons. Some of these episodes were classified by the company as biological misuse. The company presents this biological-risk focus as the most sensitive part of the report.

The avian flu case

In one case, a researcher outside the United States accessed Claude from a region where the assistant is not officially offered. At the time, he was using the model while researching highly pathogenic avian influenza, bird flu. The work partly involved studying how the virus adapts to mammals.

Anthropic classified this research as clearly dual-use, a term used when the same knowledge can help science and at the same time be dangerous if misused. In practice, understanding adaptation helps scientists identify natural variants with pandemic potential, but it can also generate information that could make a virus more dangerous. The company banned the accounts linked to the cases, but added a caveat: this does not prove the researchers wanted to cause harm.

We do not assert that they intended harm,

Anthropic, in the report

Warning amid push for new rules

This is not the first warning of its kind from the company. Anthropic had already published three threat assessments on malicious uses of artificial intelligence. The new report expands that series with a specific focus on biological risk and on disrupting uses deemed risky.

The disclosure came days after Jacob Coxon, a former Anthropic and OpenAI researcher, spoke publicly about fears over increasingly powerful systems. The debate comes as calls grow on Capitol Hill, the seat of the United States Congress, for more government oversight and new safety rules for artificial intelligence development.

If you extrapolate into the future, the level of capabilities of these AIs...They could cause extreme havoc, for example, hacking critical infrastructure, building extinction-level bioweapons,

Jacob Coxon, former Anthropic and OpenAI researcher

For those who use AI tools, the message is direct: research involving dangerous viruses is now on the list of what companies want to block inside the chat. The case also shows how enforcement works in practice, with blocked access in an unsupported region and bans on accounts linked to the cited episodes.