Anthropic said it suspended multiple user accounts after finding instances where its artificial-intelligence models were used for tasks that could have enabled the development of biological weapons. The company described biological misuse as one of the "most serious risks" associated with AI models.
What the company found
In a published report Anthropic presented five case studies in which users bypassed access restrictions and used other methods to conceal the true purpose of their AI requests. The examples included requests related to so-called "gain of function" work — laboratory efforts that aim to give pathogens capabilities they did not previously possess.
Specific instances cited by Anthropic included researchers using AI to prepare a grant proposal that, if funded, would have supported gain-of-function research on the chikungunya virus. In another case, the company's Claude model was used in queries related to avian influenza research. The report also documents activity concerning orthopoxviruses (for example mpox and viruses related to smallpox) and requests about various toxins and poisons.
Scope of the review
Anthropic reviewed a 30-day period and identified roughly 35 research activities that raised concern. The company emphasized it cannot definitively determine from these examples whether the users acted with malicious intent or whether the work served lawful scientific purposes.
The report notes that the individuals referenced were active researchers, but Anthropic did not disclose the countries or research institutions involved.
Why this matters
The cases underline the dual-use nature of AI in biological research: the same tools that can accelerate drug discovery and deepen understanding of pathogens can also be misused to develop harmful biological agents.
Anthropic says its current safety measures are functioning, but warns that more sophisticated misuse attempts would require stricter safeguards and regulation.
Responses and warnings
The disclosure reinforces researchers' concerns that AI could be harmful to humanity if it advances too rapidly or without appropriate governance. Jacob Coxon, a former Anthropic employee, told CNN that projecting the capabilities of future AI systems makes it clear these technologies could cause significant destruction if misused.
The episode raises further questions about how to balance the benefits of open AI-enabled research with the need to prevent dangerous applications.
