Claude has been used to help draft bioweapons grant applications, engineer a drone swarm, and build surveillance tools for multiple governments, according to Anthropic's newest threat intelligence report.
The report covers eight months of misuse across seven categories, including cyber operations, surveillance, biological misuse, and conventional weapons development. Anthropic says researchers, including state-supported ones, prompted Claude to write grant applications involving the chikungunya virus and orthopoxvirus, the family that includes smallpox, after circumventing geographic access controls meant to block users in countries without Claude access. A Russia-based operator allegedly used Claude Code to build a full-stack autonomous drone swarm, while a separate China-based account used other Claude work tools to design electronic-warfare modules, simulating strikes on 12 targets in Taiwan. Anthropic also linked Claude to nine state-connected surveillance efforts, including profiling Uyghur populations through WhatsApp and Telegram data, and browser extensions tied to Iranian paramilitary and domestic security agencies that harvested data from 6,388 Iranians.
The report matters less for any single case than for what it says about capability diffusion. Anthropic itself argues that AI has leveled the playing field, giving lone criminals access to tools that once required state resources. That claim lands just as Anthropic and OpenAI both lobby for California's new AI oversight bills, a coincidence worth noting.
Anthropic frames this as responsible disclosure. It is also a company grading its own safety homework, using the cases it caught to argue for the regulation it helped write.