China, Iran among countries that have used AI to aid spying, Anthropic says

Anthropic, an artificial intelligence company, has blocked several attempts in recent years from scientists using its AI models for research that could potentially contribute to the development of biological weapons. The company disclosed these findings in a report published on Thursday, marking its third since March 2025 on the misuse of its AI systems.

Anthropic expressed difficulty in determining whether the research in question was intended for weapons development or legitimate medical studies, such as vaccine creation. The company opted for caution, blocking the activity rather than waiting for certainty due to the potential for harm in biological research. A notable case involved a scientist preparing a grant application concerning the Chikungunya virus, a mosquito-borne disease causing severe pain and fever.

The scientist aimed to create mutations that would enhance the virus's transmissibility and immune system evasion, testing this through multiple live animals. The company suspected this work was intended for military research, though it refrained from specifying the scientist, institution, country, or biological agents involved.

The researchers concealed the true purpose of their work to evade Anthropic's safety protocols. Subsequently, the accounts associated with this misuse were banned. In addition to the biological research cases, the report revealed attempts to use Anthropic's AI tools for conventional weapons development, including code linked to firearms, missiles, drones, and bombs, with six cases linked to users in China, Russia, and Yemen.

The report attributed the Yemen case to the Iran-backed Houthi movement and identified potential government actor involvement in China and Iran, including efforts to spy on dissidents and expat communities. Russian state media allegedly used Anthropic's Claude AI to produce content resembling independent news, including false claims about an election in Moldova.

Anthropic disclosed these practices to promote transparency, as it believes AI companies must disclose harmful uses of their services. The company emphasized that none of the cases involved their newest Claude Fable or Mythos-class models, except for one involving unauthorized copying of model capabilities. The report's release comes two days after an Anthropic researcher, Jacob Coxon, resigned over concerns about the rapid development of highly advanced AI systems.