AI company Anthropic revealed on September 11, 2026, that Iran-linked accounts used its Claude artificial intelligence model for various malicious activities between December 2025 and August 2026. This included developing targeting recommendations against US naval forces in the Middle East by collecting and analyzing publicly available information. The accounts used Claude to create targeting handbooks, identify and track naval positions, compile names of US personnel from public photographs, and research vulnerabilities in maritime satellite terminals, Cisco communications equipment, and industrial control products.
Beyond naval targeting, the same Iran-linked account also reportedly utilized Claude to develop components for an Iranian state domestic surveillance platform. This platform integrated automatic license-plate recognition with mobile-device identifier interception. Additionally, the AI model was used to analyze social connections within a private Telegram group of 244 members.
Anthropic stated it has banned the accounts involved in these operations and shared threat intelligence with government authorities regarding the US naval reconnaissance case. Indicators from the Iran state-aligned influence operations were also shared with industry and research partners. While it's unclear if the intelligence generated with Claude was used in any specific Iranian military operation, the disclosure follows recent incidents where Iranian forces targeted US naval assets and facilities in the Gulf region, though US officials reported no casualties.
