Anthropic bans scientists after detecting covert bioweapon research on Claude
Anthropic has just revealed a number of case studies where they detected concerning uses of Claude, and one involved a group of international scientists that were seemingly using the AI tool to develop a covert bioweapon.
The threat was first discovered by Anthropic in May 2026 through the company's biological safety classifier, as it blocked a request to use Claude in authoring a grant application for scientific funding.
This research specifically related to what's referred to in the science world as 'gain-of-function', which denotes the process where an organism is genetically altered to become a new or enhanced version of the original property.
Not only was this threatening in isolation – as the scientists suggested the use of this new pathogen to spread throughout the world in a way that's hard to detect – but it also outlines the immediate dangers that threaten the world when AI tools are able to be used to create bioweapons.
Claude was nearly used to create a bioweapon
Revealed by Anthropic's threat intelligence report this month, the disease that these scientists were trying to adapt with Claude was the chikungunya virus, which is typically transmitted from mosquitos to humans.
Advert
While fatalities are rare from the standard form of the virus – occurring usually only in young babies or elderly individuals with existing health complications – chikungunya is usually associated with a fever and severe joint pain that can often be debilitating, according to the World Health Organization (WHO).
Anthropic explains that "because chikungunya circulates naturally, a deliberate release (as part of a bioweapon) would be difficult to distinguish from a natural outbreak," making this seemingly targeted development even more frightening.
"The grant sought to identify enhancing mutations in the chikungunya virus, engineer them into infectious clones, and select for virulence in vivo," Anthropic illustrates.
"In other words, the virus would become progressively more harmful as it repeatedly infected live animals, with researchers keeping the most disease-causing variants in each round.
"Similar research could certainly be used in the development of better vaccines and therapeutics for the virus — but it could also be used to make the pathogen more dangerous."
How did Anthropic deal with the scientists?
One of the biggest 'red flags' for Anthropic when assessing the viability and morality of this research was the institutional affiliation, as while the team behind it were civilian researchers, the grant indicated that it was to be performed at a military research institute — and you can see why that would set off alarm bells.
Additional investigations discovered that the researchers were in markets where Claude was not available, meaning that they had accessed the tool through gray market resellers and synthetic accounts, tunnelling traffic through U.S. infrastructure to avoid regional blocks and using a zero data retention (ZDR) service to hide the content they were working with.
"At the conclusion of our investigation in May 2026, we banned all associated accounts, worked with partners to take down the relay networks that evaded regional blocks, and shared our findings with affected AI labs and government authorities," Anthropic proclaimed.
This seemingly wasn't enough though, and within weeks the operator had already re-established access to Claude and was running platform development through ZDR partners, with Anthropic continuing to ban accounts associated with the ongoing activity.
The company asserts that it is "continuing to incorporate our investigative findings into new measures to detect and prevent the actors from misusing Anthropic's services," but it shows quite how dangerous both the access and results produced by these AI tools are at the moment, let alone in the near future.