Anthropic Blocks Malicious AI Use Linked To Weapons Development
By Kileuna Haralu ·
Anthropic has blocked attempts to misuse its Claude AI models for activities linked to biological and conventional weapons development, uncovering five cases and six software systems involved.
In its latest threat intelligence report, covering December 2025 to August 2026, Anthropic has highlighted the increasing threats involving artificial intelligence being used beyond conventional cybercrime. The company identified third parties using Claude in attempts at cyber espionage, surveillance, scams, fraud, and influence operations, raising concerns about how AI companies will prevent such malicious activities while advancing legitimate scientific and technological research.
The company described biological misuse as one of the most serious risks associated with frontier AI, while highlighting both the benefits and dangerous outcomes of its applications.
The company had reported six cases involving Claude being used as a tool to develop software for conventional weapons, including firearms, missiles, armed drones and bombs, as well as targeting and manipulating systems. Anthropic said a Yemen-based weapons engineering group used Claude to assist with missile-development work, including missile guidance systems.
Despite the concern, Anthropic has drawn a line at fully autonomous weapon systems.
"Frontier AI systems are simply not reliable enough to power fully autonomous weapons. We will not knowingly provide a product that puts America's warfighters and civilians at risk," said Dario Amodei, Anthropic CEO, defending the company’s decision not to remove safety guardrails for battlefield automation.
A case concerning the incident involved hackers whose activities were consistent with Russia-based Midnight Blizzard, also known as APT29 and Cozy Bear, associated with cyber-espionage campaigns. The group used AI to develop malware that detected attempts by security systems to identify its activities, then modified the code to evade detection.
Anthropic said it is working to improve its systems to prevent, detect, and disrupt similar activities in the future. The company claimed that it shared relevant information with authorities and industry partners. The company has also introduced stronger protections for biological research in newer models, including Claude Fable 5.