Anthropic says it blocks attempts to use AI for biological weapons work

Anthropic said it identified and disrupted five cases in which its Claude AI models were used in ways that could support biological weapons development, highlighting the growing difficulty of distinguishing legitimate scientific research from potentially harmful activity.

The cases were detailed in Anthropic's latest threat intelligence report, published on Thursday (10 September), covering malicious or potentially harmful uses of Claude between December 2025 and August 2026, reports BBC.

Anthropic said biological misuse is among the most serious risks posed by frontier AI models, warning that without appropriate safeguards, advanced AI capabilities could have "catastrophic consequences".

The company's Threat Intelligence team identified and disrupted activity across seven areas during the eight months, including cyber operations, influence operations, surveillance, scams and fraud, biological misuse, conventional weapons development and illicit model distillation.

The biological-misuse cases involved researchers using Claude for dual-use work, including virology, toxins and other areas of biological research. Anthropic said it was withholding the names of the researchers, institutions, countries, biological agents and specific techniques because the individuals were working scientists and the company had not established that they intended harm.

Five biological-misuse cases

One case involved a reseller platform that helped researchers working on state-sponsored chikungunya research circumvent regional restrictions and access AI models with weaker safeguards. Anthropic said the research involved gain-of-function work related to the virus's transmissibility and immune-evasion properties.

In another case, a researcher outside the US spent weeks using Claude in work involving highly pathogenic avian influenza and its adaptation to mammals.

Anthropic said its classifiers ultimately confined the activity to its weakest models.

A third case involved a reseller relay serving more than a dozen customers, through which Claude was used to draft an orthopoxvirus immune-evasion grant application in about an hour, Anthropic said.

Two further cases involved research into novel venoms and toxins. In one, Anthropic said a researcher's work was linked to a state-supported programme. In the other, a researcher used Claude to computationally redesign several toxins while deliberately keeping their identities vague in progress reports.

Anthropic said it banned accounts associated with the identified activity and incorporated its findings into its safeguards, enforcement systems and threat-intelligence processes.

AI's biological research dilemma

Anthropic said the cases demonstrate the challenge of policing biological research because the same scientific knowledge can have both beneficial and harmful applications.

"The same information that can be used to develop a biological weapon could also be used to develop, for example, a vaccine or a cure for a disease," the company said.

The company said older Claude models had been assessed as falling well below the threshold for meaningfully assisting sophisticated users with dangerous biological research. But it said the evidence is less certain for newer models capable of supporting increasingly complex scientifi