They tried to use Claude for research related to biological weapons

Компания Anthropic заявила, что заблокировала деятельность неназванных пользователей, которые пытались использовать ее модели искусственного интеллекта для исследований, способных привести к созданию биологического оружия, передает Kazinform со ссылкой на BBC.

This is just one of the identified cases of malicious use of Claude's AI models.

“The examples we describe are not typical cases of [models] abuse, but rather examples of the most prominent and novel threat activity we have identified to date,” Anthropic said in a report on AI abuse published September 10.

“The examples we describe are not typical cases of [models] abuse, but rather examples of the most prominent and novel threat activity we have identified to date,” Anthropic said in a report on AI abuse published September 10.

The company's report provides examples of code and AI prompts that potentially pose a threat. Anthropic called on governments and other companies developing AI to find and prevent such abuse of models.

“We are publishing this report because we believe we have a responsibility to uncover malicious abuse of our services,” the company explained.

“We are publishing this report because we believe we have a responsibility to uncover malicious abuse of our services,” the company explained.

Anthropic researchers found abuse of the models between December 2025 and August 2026 by spyware developers, “politically motivated individuals,” and state-sponsored propaganda groups.

Five examples in the report concern the potential development of biological weapons. Among them is a request for assistance in writing a research grant application.

In it, the report says, user Claude wanted to apply for research that would genetically modify the chikungunya virus. It has been proposed to change the properties of the virus, such as its infectivity and ability to bypass the immune system. Anthropic notes that while such research could help develop vaccines and drugs, “it could also be used to make a pathogen more dangerous.”

Just one example in the company's report concerns their latest models. Anthropic calls it "an industrial-scale covert campaign to figure out a model's abilities and replicate them without permission." Older models, the company says, could not significantly help the user “conduct dangerous biological research,” so the safety precautions were less stringent.

Earlier, an Anthropic expert named the likelihood of a catastrophe due to AI in the next 10 years.