An Iran-linked group used Anthropic’s Claude to collect and analyze publicly available information and develop targeting recommendations against U.S. naval forces, according to a new Anthropic threat-intelligence report.
The adversary used Claude to help build a Python pipeline that compiled targeting handbooks and identified and tracked naval positions using open-source information. The material included publicly available ship and aircraft transponder identifiers, commercial satellite-imagery queries and U.S. personnel information scraped from captions on public military photographs.
The actor also directed Claude to research vulnerabilities in shipboard systems, including maritime satellite communications terminals, Cisco communications equipment and industrial-control products.
Anthropic said it banned the account, developed additional safeguards and shared intelligence with government authorities. The company described the actor as having an Iran connection but did not identify the group or establish in the report that the targeting work resulted in an attack.
The operation was one of several cases Anthropic detailed involving attempts to use Claude for military purposes. A group in northern Yemen used Claude to develop guidance, navigation and control software for weapons programs that included a multistage ballistic missile. Anthropic said its safeguards blocked many requests from the group, “but not all of them.”
In another case, Russia-based actors used Claude to develop software for an autonomous military drone swarm, including a computer-vision system trained on Ukrainian combat footage. Anthropic also identified several China-based operations that used Claude for surveillance and intelligence activities, including tracking and profiling Uyghurs and compiling dossiers on religious leaders and diaspora figures across Asia.
Anthropic separately said it disrupted five cases in which its models were used for biological research that could potentially support biological weapons development. The company did not identify the researchers or institutions involved and said it was not asserting that the scientists intended harm.
The report covers malicious or potentially dangerous activity Anthropic detected and disrupted from December 2025 through August 2026.