Anthropic Blocks State-Sponsored AI Attempts to Build Bio-Weapons and Malware

Anthropic released a threat intelligence report on September 11, 2026, detailing systemic efforts by state-sponsored actors to misuse its AI models. Between December 2025 and August 2026, the company identified and blocked attempts to leverage its technology for biological weapons research, cyberattacks, surveillance, and influence operations.

Findings show threats originated from a diverse array of high-stakes actors, including an Iranian propaganda institution and Chinese AI firms attempting to replicate the capabilities of the Claude model. The report specifically highlighted the Russia-based cyber espionage group Midnight Blizzard. In one instance, a group consistent with Midnight Blizzard used AI to develop a system capable of automatically rewriting malware code to bypass security defenses.

Biosecurity was a primary concern in the report. Anthropic blocked a request to help author a grant application for gain-of-function research on the chikungunya virus, which sought to enhance the virus's immune evasion and transmissibility. Jacob Klein, Anthropic's Head of Threat Intelligence, noted that while biological misuse is a serious risk, these threats rarely mirror cinematic tropes. He stated that misuse is not typically a "comic book" scenario where an actor explicitly expresses a desire to kill everyone.

This disclosure arrives during a period of escalating tension regarding AI safety and the speed of development. The report coincides with the resignation of Jacob Coxon, a former Anthropic researcher who warned that the current race toward superintelligence is "gambling with our lives." Simultaneously, US Senator Bernie Sanders has called for a ban on superintelligence and a pause on the development of advanced AI systems to mitigate existential risks.

These findings highlight the "dual-use" nature of frontier AI, where tools designed for medical breakthroughs, such as vaccine development, can be inverted to enhance pathogens. Anthropic described biological misuse as one of the most serious risks facing the industry, warning that without stringent safeguards, such capabilities could lead to catastrophic consequences.

As models become more capable, the report suggests that voluntary safety commitments may be insufficient. The active targeting of frontier models by state-sponsored criminals and intelligence agencies underscores the need for non-voluntary, systemic safeguards. Industry observers are now watching whether these findings will trigger more aggressive regulatory frameworks or if the competitive pressure to reach superintelligence will override the safety warnings issued by both internal researchers and government officials.