Anthropic says it blocked misuse of its AI that could have supported biological weapons

Anthropic, a leading AI company, has uncovered and thwarted attempts to misuse its AI models for malicious activities, including the development of biological and conventional weapons, according to the company's latest threat intelligence report. The report, published on Thursday, details cases from December 2025 to August 2026 where suspected state-sponsored groups, criminals, spyware vendors, state propaganda institutions, and politically motivated individuals abused Anthropic's AI technology.

Between December 2025 and August 2026, Anthropic detected malicious use of its Claude Haiku, Sonnet, and Opus models in various contexts. These included cyber and influence operations, surveillance, scams, fraud, and weapon development. However, none of the misuse cases involved the Claude Fable or the powerful Mythos-class models, with a single exception of distillation, the process for training smaller AI models using larger, more expensive models.

The report highlights five case studies of actors using Anthropic's models in ways that could support biological weapons development, emphasizing that such misuse is "one of the most serious risks of frontier AI models." The company warns that the same information that can be used to create a biological weapon could also be used to develop a vaccine or cure for a disease, underscoring the potential catastrophic consequences if safeguards are not in place.

Anthropic's head of threat intelligence, Jacob Klein, stressed that the misuse of Claude is a nuanced issue, and the company has not encountered cases of individuals attempting to build biological weapons with the sole intent of causing widespread harm. However, the report does note six instances where Claude was used to develop software for conventional weapons, such as firearms, missiles, armed drones, bombs, and targeting and control systems.

Cybercriminals and state-backed hackers have increasingly leveraged Anthropic's AI technology to support their operations, as evidenced by cases involving hacking group ShinyHunters and China-based labs. The report also cites a hacking group associated with the Russia-based Midnight Blizzard that allegedly used AI to create a system that automatically altered malware code to evade detection.

In response to these findings, Anthropic has incorporated its threat intelligence into its processes to enhance its ability to prevent, detect, and disrupt malicious activities in the future. The company has also shared its intelligence with relevant authorities and industry partners where appropriate, demonstrating its commitment to addressing the risks associated with AI technology.