Anthropic, a U.S.-based artificial intelligence company, has reported multiple attempts this year to misuse its AI model Claude for research related to potential biological weapons. The company’s recently released misuse report highlights concerns that scientists may be exploiting its platform to pursue projects that pose significant risks to public safety.
According to the report, Anthropic identified efforts by researchers who sought to bypass the company’s safeguards designed to prevent access by users from restricted regions or those with questionable intentions. These individuals allegedly took steps to conceal the true nature of their inquiries, aiming to evade detection and restrictions. In response, the company has banned the accounts involved.
The report stops short of specifying the identities of the researchers, their affiliated institutions, or their countries of origin. It also does not detail which biological agents were involved. However, Anthropic noted in one documented case that a scientist requested Claude’s assistance in drafting a grant proposal for experimental research on the chikungunya virus, a mosquito-borne illness known for causing high fever and severe joint pain. The proposal was reportedly linked to a military research institute, raising additional concerns about the potential application of the research.
Anthropic’s officials acknowledged the challenge of distinguishing between legitimate biological research—which can lead to important breakthroughs such as vaccine development—and activities that might facilitate the creation of dangerous pathogens. The company opted to prioritize caution, emphasizing the potential risks posed by allowing such inquiries to proceed unchecked.
The revelations come amid broader debates about the risks posed by rapidly advancing AI technology. Former Anthropic researcher Jacob Coxon recently resigned from the company, expressing criticism of Anthropic and another major AI developer, OpenAI, over perceived insufficient measures to address the threats associated with AI. Coxon warned of a more than 10 percent chance that AI could contribute to human extinction within five years, reflecting growing unease within the research community.
Anthropic’s report outlines several cases of AI misuse detected over the past eight months, underscoring the ongoing challenges faced by AI developers in monitoring and controlling the applications of their technologies. The company has not disclosed further details, and efforts to obtain additional comment remain pending.