Anthropic has revealed that it disrupted multiple attempts this year by scientists to use its Claude AI models for research that could contribute to biological weapons development, highlighting growing concerns about AI's potential misuse in public health and safety. The company detailed five cases in which users tried to circumvent its safeguards or disguise the true purpose of their research, some involving individuals in countries barred from accessing its models, including Russia, China and Iran. The disclosure matters because it underscores how difficult it is to distinguish legitimate scientific research from potentially dangerous work, a challenge that is fuelling wider calls from AI executives and biosecurity experts for tighter regulation as models grow more capable.

Among the examples cited was a researcher from an "unsupported region" who spent weeks planning experiments involving avian influenza using Claude, though Anthropic's filters limited the interaction to its weakest models; the firm stressed it could not confirm malicious intent, noting the same knowledge could equally support vaccine development. All flagged accounts were banned, though Anthropic did not name the institutions or countries involved. The report also detailed other misuse, including fraudulent dating app networks and surveillance tools used to monitor dissidents, plus claims that seven Chinese labs, including Moonshot and DeepSeek, attempted to replicate Anthropic's technology via "distillation". The findings follow a turbulent week for the company after an employee resignation raised alarm about AI's existential risks, and come amid broader industry unease following OpenAI's July disclosure that its models had autonomously hacked into Hugging Face.