Anthropic Says It Blocked 5 Attempts to Misuse Claude for Potential Bioweapons Research

If you thought AI misuse was mainly about scams or cyberattacks, Anthropic’s latest report is all about a far more serious concern. The company says Claude was targeted in five cases involving biological research that could potentially support bioweapons development.

Claude was used for potentially dangerous biological research

Anthropic disclosed the cases in its latest threat intelligence report, covering activity between December 2025 and August 2026. The company said it blocked all five requests and banned accounts connected with the activity.

The cases involved research connected to pathogens including chikungunya and avian influenza. Anthropic stressed that it could not always determine whether researchers had harmful intentions. That’s important because biological research can have legitimate medical uses. The same research can also potentially make pathogens more dangerous, creating a difficult safety problem for AI companies.

One case involved a request connected to gain-of-function research on chikungunya. Anthropic said the research focused on properties involving transmission and immune evasion. The company also described attempts to bypass restrictions through reseller platforms and regional workarounds. Some users reportedly split their activities across multiple sessions to conceal their broader objectives.

Anthropic is bolstering Claude’s biological safeguards

Anthropic says newer Claude models are substantially better at complex scientific work than earlier versions. That progress has forced the company to rethink how biological queries are handled. The company now applies stronger safeguards to newer models, including Claude Fable 5. These controls restrict access to a broader range of dual-use biological research requests.

At a time when the latest AI models are becoming more advanced in automation and scientific research, this issue could become harder to manage. Anthropic’s findings suggest safeguards may need constant updates as AI capabilities improve.

The company also shared its findings with authorities and industry partners where appropriate. That being said, it remains unclear how quickly other AI companies can detect similar activity. Let’s not forget that ex-Anthropic researcher recently sent out a chilling warning saying that AI has over 10 percent of chance of killing all humans within next decade.