Anthropic says Iran-linked operators used Claude AI to map US naval assets

US artificial intelligence company Anthropic has said an Iran-linked account used its Claude AI system to gather and analyse publicly available information on US naval assets and develop detailed recommendations that could potentially assist in targeting them.

According to Anthropic, the operators used Claude to compile what it described as "targeting handbooks", while also tracking the locations and movements of US naval assets through information available from open sources.

The company said the account built a Python-based pipeline with Claude's assistance to collect and monitor publicly available data on naval positions. The information was then organised into material that could provide a consolidated picture of US military assets and their movements.

The material reportedly included the names of US military personnel extracted from captions accompanying publicly available military photographs, as well as ship and aircraft transponder identifiers. The operators also used scripts to query commercial satellite imagery and compiled lists of websites that publish information about US naval movements.

Anthropic said Claude was further used to examine potential vulnerabilities in shipboard systems, including maritime satellite terminals, Cisco communications equipment and industrial control products. The operators also used the AI model to catalogue publicly known Common Vulnerabilities and Exposures (CVEs) affecting maritime VSAT terminals.

The company said it had identified and disrupted the account after detecting the activity. However, Anthropic acknowledged that it remains unclear whether intelligence generated through Claude was ultimately used in any specific Iranian military operation.

The disclosure highlights a growing concern surrounding the use of generative AI in military and intelligence activities. Publicly available information that is individually harmless can potentially become more operationally significant when AI systems are used to rapidly collect, organise and analyse large quantities of data.

Anthropic's report also identified another account linked to Iran's Islamic Culture and Communications Organization. That account reportedly uploaded detailed organisational material concerning the funeral arrangements and succession planning associated with former Iranian Supreme Leader Ayatollah Ali Khamenei.

Khamenei died in a US-Israeli strike on February 28, 2026, and was buried in Mashhad on July 9 following a multi-day state funeral procession.

The report comes amid heightened tensions between Iran and the United States, including attacks involving US military assets and facilities in the Gulf region. US officials have said that American warships were able to evade attacks without casualties, although a facility at Naval Support Activity Bahrain was damaged.

Anthropic's findings underline how AI tools designed for general-purpose information processing can potentially be repurposed for military intelligence work, particularly when combined with extensive open-source information and automated data-analysis tools.