Anthropic catches scientists covertly using Claude for lethal bioweapons research

Anthropic says it disrupted five cases in which researchers used Claude in ways that could have supported biological weapons development. The company said it could not establish whether the researchers intended to cause harm. It still blocked the activity because the potential consequences were too serious to ignore. The cases surfaced between December 2025 and August 2026. Anthropic described them in a new threat intelligence report covering misuse of its models. The company said the incidents show how harder-to-detect biological risks can emerge as frontier models become better at scientific research. The biology dilemma Anthropic’s biggest challenge was determining where legitimate research ended and dangerous work began. Biological research often has dual-use potential. The same scientific knowledge can support medical treatments or enable more dangerous pathogens. Jacob Klein, Anthropic’s head of threat intelligence, told The New York Times that the situation rarely involves obvious declarations of malicious intent. Researchers can present seemingly legitimate scientific goals while pursuing work that raises serious security concerns. We're publishing our most detailed threat intelligence report to date. It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.We disrupted every operation in the report,…— Anthropic (@AnthropicAI) September 10, 2026 Anthropic said its investigation found attempts to bypass regional restrictions and evade safety systems. It banned accounts linked to the activity and shared findings with authorities and other AI companies. One case involved research linked to chikungunya, a mosquito-borne virus. Anthropic’s systems blocked a request involving a grant application for gain-of-function research. The associated work was intended for a military research institute, which increased the company’s concern. The researchers continued their work through other channels after Claude blocked sensitive requests. Anthropic also found evidence that a reseller platform routed refused biology prompts toward models with weaker safeguards. Frontier models raise stakes A second case involved research into highly pathogenic avian influenza. Anthropic said a researcher spent weeks using Claude for study planning, data analysis and scientific interpretation. The company said its safeguards prevented the researcher from accessing its strongest models for the sensitive work. The activity instead remained on weaker model versions. Anthropic reported three additional cases involving biological research with potential weapons implications. Those cases included work involving orthopoxviruses, venom-related compounds and toxins. The company withheld the researchers’ identities, locations and specific biological details. Anthropic said it did not claim the scientists intended to cause harm. The distinction matters because older Claude models showed limited ability to support sophisticated biological research. Anthropic now says it cannot make the same assumption about newer systems. Susan Monarez, a microbiologist and former U.S. public health official, reviewed the report before publication. She told The New York Times that advanced models could help malicious actors hide dangerous research behind legitimate scientific objectives. Anthropic has responded by tightening controls around high-risk biological queries. The company also says it wants researchers to retain access to useful scientific capabilities.That creates a difficult engineering problem. Safeguards must block dangerous assistance without crippling legitimate research into diseases, vaccines and treatments. Anthropic’s latest cases suggest that distinction is becoming harder as frontier models gain deeper scientific capabilities.Get the latest in engineering, tech, space & science - delivered daily to your inbox.Aamir is a seasoned tech journalist with experience at Exhibit Magazine, Republic World, and PR Newswire. With a deep love for all things tech and science, he has spent years decoding the latest innovations and exploring how they shape industries, lifestyles, and the future of humanity.