Anthropic Blocks Accounts Over Suspected Bioweapons Research

Anthropic said Thursday it had blocked accounts that it believed may have been involved in efforts to use its Claude AI model in research that could potentially contribute to the development of biological weapons. The company disclosed the findings in an investigation released days after AI researchers raised broader concerns about the possibility of increasingly capable AI systems posing an existential threat to humanity.

According to Anthropic’s report, the company stopped a request seeking Claude’s help in preparing a grant application for scientific funding related to genetic modification research involving chikungunya, a mosquito-borne virus. Anthropic said the proposed work appeared aimed at altering the virus in ways that could make it more harmful.

The company acknowledged that similar genetic research could have legitimate applications, including the development of vaccines or treatments for chikungunya. However, Anthropic said the potential for the work to increase the virus’s danger raised concerns about how the research could ultimately be used.

Anthropic said it became particularly concerned because the grant materials indicated that civilian researchers would be involved, while the proposed research itself was expected to take place at a military research institute. The company said this discrepancy contributed to its assessment that the project may not have been as benign as the funding proposal suggested.

The company has not identified the institutions or researchers that may have been connected to the activity. Anthropic said it banned all accounts associated with the suspected operation, dismantled relay networks that had been used to circumvent its regional restrictions and provided its findings to other AI laboratories and the federal government.

Anthropic said the materials it uncovered suggested the activity extended beyond simply preparing a funding application. “We infer from the existence of these research materials that the research effort was not limited to a funding proposal,” the company said. Anthropic added that it was banning accounts linked to the activity and using the findings from its investigation to strengthen systems designed to identify and prevent misuse of its services.

The disclosure comes as concerns over the risks of advanced AI have intensified. This week, former OpenAI and Anthropic staffer Jacob Coxon warned that AI development could potentially lead to human extinction within the next decade. Evan Hubinger, who leads Anthropic’s efforts to align AI systems with human goals and values, separately said he personally believes there is a more than 10% chance that AI could “kill all humans” within the next 10 years if development continues to accelerate.

Hubinger made that assessment in response to Coxon, who recently left Anthropic after a short period at the company. Coxon accused AI companies of behaving irresponsibly by “racing straight to self-improving superintelligence and gambling with our lives.” He also said researchers in the field believe AI could “kill us all by the end of the decade” and insisted that his warning was not a “marketing stunt.”

The warnings from Coxon and Hubinger have prompted calls from lawmakers for stronger restrictions on AI development. They have included proposals for an “AI Kill Switch Bill,” while Sen. Bernie Sanders, I-Vt., has said he intends to introduce legislation that would ban superintelligence and impose a pause on AI development.

President Donald Trump, meanwhile, has strongly supported continued AI development in the United States. His administration has encouraged communities to accommodate data center construction and backed the broader competition between the U.S. and China to advance artificial intelligence.