Scientists used Claude to explore ways to make a mosquito-borne virus more infectious, while other users sought help with weapons development and surveillance

Anthropic has revealed that it blocked several attempts to use its artificial intelligence models for potentially harmful biological research, weapons development and state-linked surveillance.

The US-based AI company said on Thursday that it had identified multiple users trying to use Claude, its AI chatbot, for activities that could pose serious security risks.

In one case, an unnamed scientist sought help in May to design experiments involving chikungunya, a mosquito-borne virus that can cause severe joint pain and other debilitating symptoms.

The scientist was researching ways to make the virus more infectious and potentially more harmful. While similar research can be conducted for legitimate purposes, such as developing vaccines, Anthropic said it became concerned after learning that the work was reportedly going to be carried out at a military research institute.

Anthropic said it banned all accounts linked to the activity and shared information about the case with relevant government authorities. It did not identify the country involved.

The company described biological misuse as one of the most serious risks associated with AI. It said the cases it uncovered represented a significant warning about how increasingly capable AI systems could be used to support dangerous research.

Anthropic also reported attempts to use Claude to develop firearms, ballistic missiles, armed drones and bombs.

Three of those cases were linked to users in China, two to Russia and one to Yemen, which Anthropic said was believed to be connected to the Iran-backed Houthi militia.

The company also said it had prevented Iran from using Claude to identify US naval bases and individual military personnel.

In another case, Anthropic said an account operating on behalf of Iran’s Islamic Culture and Communications Organization uploaded detailed plans related to the funeral of former supreme leader Ali Khamenei and the country’s leadership succession.

The report comes as concerns grow over the use of AI in military operations and the potential consequences of increasingly powerful models.

Anthropic itself has faced scrutiny over the military use of its technology. The Pentagon said earlier this year that it had used Anthropic’s AI to produce intelligence assessments, identify targets and simulate battles during US military operations against Iran.

Anthropic chief executive Dario Amodei said in February that the company could not “in good conscience” agree to the US government’s demands for unrestricted and unconditional military use of its AI.

The company’s latest report also documented alleged surveillance activities involving Russian and Chinese accounts.

Anthropic said one Chinese state-linked operation used Claude to track and profile Uyghurs and armed Uyghur groups in Syria. Another operation involving Chinese municipal police reportedly sought intelligence on anti-Beijing protests in Canada and attempted to track individual dissidents, including Chinese artist and activist Teacher Li, who lives in exile in Italy.

The revelations come amid growing calls from AI researchers for greater caution over the rapid development of the technology.

Jacob Coxon, a former OpenAI researcher who recently resigned from Anthropic, said neither company was acting responsibly.

Coxon’s resignation has added to concerns in Silicon Valley over the potential risks posed by increasingly advanced AI systems, with the researcher warning that the industry was “gambling with our lives” as development accelerates.