- Anthropic blocked five instances of researchers using Claude to support bioweapons development.
- Distinguishing between legitimate research and dangerous work is challenging, as biological research has dual-use potential.
- Anthropic released a detailed threat intelligence report describing various ways Claude was misused.
- As model capabilities advance, distinguishing between legitimate and dangerous research becomes increasingly difficult.
Anthropic has blocked five instances of researchers using Claude to potentially support bioweapons development. The company states it cannot confirm malicious intent but halted the activities due to the severe potential consequences. These cases occurred between December 2025 and August 2026, and Anthropic described the misuse of these models in a new threat intelligence report. The company said these incidents show that as frontier models advance in scientific research capabilities, less detectable biological risks may emerge.
The Boundary Between Legitimate and Dangerous Research
Anthropic's biggest challenge lies in distinguishing between legitimate research and dangerous work. Biological research often has dual-use potential. The same scientific knowledge can support medical treatments or facilitate more dangerous pathogens. Jacob Klein, Anthropic's head of threat intelligence, told The New York Times that such cases rarely involve explicit declarations of malicious intent. Researchers can present seemingly legitimate scientific goals while engaging in work that raises serious security concerns.
"We are releasing the most detailed threat intelligence report to date. It covers how people have tried to misuse Claude—for cyberattacks, influence operations, surveillance, biological research, and weapons construction—as well as how we detected and stopped these activities," Anthropic said on social media. "We stopped every operation detailed in the report."
Case Studies of Claude Misuse
Anthropic said its investigation uncovered attempts to bypass regional restrictions and evade security systems. The company banned accounts associated with this activity and shared its findings with authorities and other AI companies. One case involved research related to chikungunya, where Anthropic's system blocked a request for funding related to gain-of-function studies. The related work was intended to be conducted at a military research facility, which heightened the company's concerns. After Claude blocked the sensitive request, the researchers continued their work through other channels.
Anthropic also found that a resale platform was redirecting rejected biological prompts to models with weaker safety protections. A second case involved research on highly pathogenic avian influenza. Anthropic said a researcher spent weeks using Claude for research planning, data analysis, and scientific interpretation. The company noted that its security measures prevented the researcher from accessing its most powerful models for sensitive work, and the activity ultimately remained on a weaker model version.
Control Measures for High-Risk Biological Queries
Anthropic also reported three other cases involving biological research with potential weapons implications. These cases included work related to orthopoxviruses, toxins, and venom-related compounds. The company withheld the researchers' identities, locations, and specific biological details. Anthropic said it does not claim the scientists intended to cause harm. This distinction is crucial because older Claude models showed limited ability to support complex biological research. Anthropic now says it cannot make the same assumption about the new system.
Microbiologist and former U.S. public health official Susan Monarez reviewed the report before its release. She told The New York Times that advanced models might help malicious actors hide dangerous research behind legitimate scientific goals. Anthropic has strengthened control measures for high-risk biological queries. The company also said it hopes researchers can retain access to useful scientific capabilities, creating a difficult engineering problem. Security measures must stop dangerous assistance without hindering legitimate research on diseases, vaccines, and treatments.
Anthropic's recent cases show that as frontier models gain deeper scientific capabilities, this distinction is becoming increasingly difficult.
The Impact of Frontier Models on Biological Research
As AI models advance, researchers may conduct potentially dangerous biological research under the guise of legitimate scientific goals. Anthropic's report reveals this phenomenon, showing that the potential risks of dual-use in the field of biological research are increasing. This not only challenges the security measures of tech companies but also raises questions about how to balance scientific progress with public safety. As technology rapidly evolves, ensuring research is not misused is becoming an important challenge, with far-reaching implications for future scientific research and its applications.