Anthropic Identifies User Planning Bird Flu Research

A researcher used the Claude chatbot to design experiments for adapting avian influenza for mammals.

Updated on Oct. 26, 2026 in Artificial Intelligence

Isometric editorial illustration of a glass laboratory vial on a white surface, evoking technical security and biological research policy. — Anthropic identified a user attempting to use its Claude AI to plan high-risk avian influenza research in May 2026. AI Illustration. Upload story photo >

Live Poll

Should government place stricter regulations on how AI models assist in potentially risky biological research?

In May 2026, AI company Anthropic identified a user who employed its Claude chatbot to plan dangerous research involving highly pathogenic avian influenza. The researcher sought to identify mutations that could help the virus adapt to mammals.

Why it matters

The case highlights emerging security risks as AI tools become capable of assisting in the design of biological weapons. Anthropic identified five total instances of Claude being used in ways that could support such development.

The researcher utilized Claude to aggregate research papers, synthesize study results, and parse complex biological data. Anthropic reported five separate instances where its chatbot was leveraged for activities potentially aligned with biological weapons development.

The players

Anthropic

This is an AI safety and research company that develops the Claude large language model.

The details

The user bypassed geographic restrictions to access the chatbot and used it to select experimental protocols for bird flu. The material provided to the model suggested the individual possessed physical access to virus samples.

2012: H5N1 studies sparked global safety debates.
May 2026: Anthropic banned the user for restricted access.
September 10, 2026: Anthropic released its threat report.

The Tech Race

This development underscores the transition from traditional academic research risks to AI-augmented threats in the biological sciences. It mirrors the precedent set by the 2012 H5N1 bird flu safety debate, suggesting that digital tools now play a primary role in how research is designed and executed.

While this incident highlights backend security concerns for AI providers, it underlines the increasing scrutiny on user prompts involving sensitive scientific data. Researchers and developers should expect stricter implementation of safety guardrails for queries related to high-risk biological topics.

The takeaway

AI companies are actively monitoring for instances where chatbots are used to circumvent biological safety standards. Users should be aware that unauthorized experimentation requests are tracked and can lead to immediate account termination.