Altering bird flu, chikungunya: Was Claude tasked to weaponise viruses?
Artificial intelligence company Anthropic has disclosed a series of alarming cases in which researchers attempted to use its Claude AI models for biological research that could potentially support the development of dangerous pathogens, toxins and other biological weapons.
In its latest threat intelligence report, Anthropic detailed five cases involving scientists and researchers working on highly sensitive areas of bioscience.
The company stressed that it does not accuse the individuals involved of intending to create biological weapons, but said the cases demonstrate how frontier AI systems can increasingly be used to accelerate research with both beneficial and potentially catastrophic applications.
The report marks one of the most detailed public disclosures yet by an AI company about real-world attempts to use advanced models in potentially dangerous biological research. The cases demonstrate how frontier AI systems can increasingly be used to accelerate research. (Photo: Getty)
Anthropic said biological misuse is among the most serious risks associated with increasingly capable AI systems.
While older models were considered unlikely to provide meaningful assistance to sophisticated biological weapons programmes, the company said newer systems can assist with increasingly complex scientific tasks, making the distinction between legitimate and malicious research more difficult.
One of the most concerning cases involved research into the chikungunya virus, a mosquito-borne disease that can cause severe fever and debilitating joint pain.
Anthropic said its safety systems blocked a request involving a grant application for research designed to examine changes related to the virus's transmissibility and ability to evade immune responses.
The proposed work involved gain-of-function research and was linked to a military research institute, although the application suggested that civilian researchers were involved. Anthropic said the research could potentially contribute to vaccine and therapeutic development but could also be used to make the virus more dangerous.
The company also uncovered a researcher using Claude to assist work involving highly pathogenic avian influenza and its adaptation to mammals. The research explored mutations associated with mammalian adaptation and airborne transmission in animal models.
Anthropic said the researcher used the AI system over several weeks for study planning, data analysis and interpretation of experiments. However, its safety systems restricted access to more capable models, limiting the assistance largely to clerical and research-planning support.
Another case involved an application for orthopoxvirus research at a state-associated infectious disease laboratory. Claude was reportedly used to draft a complete grant proposal covering the scientific hypothesis, experimental design and other research planning. Researcher used the AI system over several weeks for study planning, data analysis and interpretation of experiments. (Photo: Getty)
Orthopoxviruses include the family of viruses that includes smallpox and mpox.
The remaining cases involved research into venom-derived compounds and computationally redesigned toxins. Anthropic said one researcher used Claude to develop a database of venom toxin peptides and a system for optimising their properties. While the stated goal involved developing painkillers and other therapies, the same research could potentially be applied to harmful or incapacitating compounds.
In another case, a researcher used Claude to assist projects involving redesigned toxins connected to a national research programme. The researcher allegedly asked the AI to keep descriptions of some biological agents deliberately vague in progress reports.
Anthropic said it banned accounts linked to the cases and strengthened its safeguards. But the report highlights a deeper challenge facing AI companies: much of modern biological science is inherently dual-use.
“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,” Anthropic said in its report.
The company warned that sophisticated actors could hide potentially dangerous objectives behind seemingly legitimate scientific research, making AI safety systems increasingly dependent not just on detecting dangerous questions, but also on understanding who is using advanced scientific capabilities and for what purpose.- EndsPublished By: Sibu Kumar TripathiPublished On: Sep 11, 2026 11:31 IST
