San Francisco, September 11 — Anthropic disclosed on Thursday that it had detected and disrupted Chinese state actors attempting to abuse its Claude AI model to develop military-grade weapons systems. The California-based artificial intelligence company identified three separate cases originating in China in which actors used the platform to develop software for conventional weapons, including missiles, armed drones and targeting systems. The disclosures form part of a comprehensive threat intelligence report covering malicious activity between December 2025 and August 2026.
Anthropic characterised the weapons cases as emerging categories of threat actors misusing Claude to develop software for firearms, missiles, bombs and the guidance and control systems that operate them. The company detected incidents involving operators from China, Russia and Yemen attempting to use Claude for weapons design, intelligence gathering and procurement related to weapons programmes.
In one documented case, a Chinese actor presenting as a US defence contractor used Claude to draft Chinese military specifications. The report detailed how actors employed Claude Code in place of software engineers to develop guidance and control code for weapons systems. Anthropic banned the accounts involved and shared intelligence with law enforcement and industry partners.
Beyond weapons
The weapons development cases represent just one category of malicious activity outlined in the report. Anthropic also blocked scientists using Claude in ways that could support biological weapons development, with five cases flagged involving gain-of-function research conducted at military institutes. The company detected Russian-linked cyber espionage campaigns, Chinese surveillance operations targeting dissidents, and what it termed industrial-scale efforts by seven China-based AI labs to illicitly replicate Claude’s capabilities through unauthorised distillation attacks.
Jacob Klein, head of threat intelligence at Anthropic, told media that AI models have become significantly more capable over the past year, raising new risks. The company said it shared relevant intelligence with law enforcement agencies and incorporated findings into its ongoing work on safeguards.
Why it matters
Anthropic bars the use of Claude for weapons design under its usage policies, though the report demonstrates how state-backed actors are circumventing restrictions to access advanced AI capabilities for military applications. The disclosure marks a significant escalation in concerns about artificial intelligence misuse for military purposes. Anthropic is an American AI safety and research company founded in 2021 by former OpenAI employees, and Claude is its family of large language models.
The company has published usage policies prohibiting weapons development, and its threat intelligence reporting adds to a growing public record of state-linked attempts to repurpose commercial AI systems for military and surveillance ends. The practical upshot for anyone following AI policy: expect more scrutiny of how commercial models are accessed, and more pressure on labs to publish what they catch.
