Anthropic Uncovers Use of Claude AI in Biological Weapons Research and Russian Espionage Operations

AI Exploitation in Hazardous Biological Studies Exposed

According to НВ — Техно: Anthropic has identified five separate instances where its Claude AI system was employed for research linked to biological weapons, leading to the suspension of the involved accounts. These incidents occurred between December 2025 and August 2026. Researchers investigating viruses, toxins, and poison-related compounds attempted to leverage the AI for conducting perilous experiments. One investigation specifically involved the chikungunya virus, with a related grant application query being blocked. Despite collaboration with a military science institution, the researchers sought alternative channels to continue their work after restrictions were imposed.

Additionally, Anthropic uncovered a proxy platform that redirected denied queries to other AI models with looser safeguards. Another case involved a researcher studying highly pathogenic avian influenza who used Claude over several weeks to plan, analyze, and interpret scientific data about the virus. The company restricted access to its most advanced AI versions for this project, forcing the researcher to rely on less capable models. The other three cases concerned:

• orthopoxviruses,
• poison-related compounds,
• toxins.

For privacy and security reasons, Anthropic has withheld the identities, locations, and detailed biological specifics of the researchers involved. All incidents have been reported to law enforcement agencies and shared with other AI developers to prevent further misuse.

Russian Cyber Espionage Campaign Linked to Claude AI

Beyond these findings, Anthropic disclosed a Russian cyber operation utilizing Claude to automate espionage activities associated with the Midnight Blizzard group. Targets included Ukrainian government and military bodies, diplomatic missions, and defense contractors across Europe. The AI facilitated phishing attacks, infrastructure setup, malware development, and data theft. Anthropic identified over 20 organizations involved in planning or executing this campaign. Their automated systems continuously monitored cybersecurity defenses to evade detection by adapting malware as needed.

In response, Anthropic has tightened restrictions on hazardous biological queries and enhanced protections across its platforms handling sensitive research.

Jacob Klein, a biological security expert, commented: "Distinguishing between legitimate scientific inquiry and potentially dangerous activities is challenging. Biological knowledge can advance medicine but also enable the creation of more harmful pathogens."

The uncovering of AI misuse in dangerous research highlights the critical need for stringent oversight of technologies with dual-use potential. As a leading AI developer, Anthropic recognizes its responsibility to prioritize public safety alongside innovation. Strengthening access controls on risky biological requests marks a vital step forward, yet ongoing vigilance and cooperation among developers and regulators remain essential to ensure AI is used safely across scientific and technological fields.

The implications of AI misuse extend beyond biological research. Recent concerns have also emerged regarding the resignation of an Anthropic researcher, who cited fears about losing control over self-improving artificial intelligence. This highlights the broader ethical dilemmas surrounding AI technologies and their potential for exploitation in various domains.