AI firm Anthropic has exposed a sophisticated operation where Iran-linked users leveraged the Claude AI model to aggregate intelligence on US Navy ship movements and identify vulnerabilities in onboard communication systems.

  • Iran-linked groups used Claude AI to transform public data into military targeting materials.
  • The operation targeted US Navy communication systems, including Cisco equipment and satellite terminals.
  • AI was utilized for multi-language propaganda campaigns and domestic surveillance systems.

In a startling revelation that underscores the dual-use nature of artificial intelligence, Anthropic has detailed how state-aligned actors from Iran weaponized its Claude AI model. Between December 2025 and August 2026, several accounts linked to Iran systematically used the AI to gather and organize intelligence on the US Navy, effectively turning a commercial productivity tool into a reconnaissance asset.

The operation involved the aggregation of vast amounts of open-source intelligence (OSINT). The actors fed Claude data from ship and aircraft transponders, commercial satellite imagery, and military photographs. The AI was then tasked with synthesizing this raw data into comprehensive targeting handbooks and tactical recommendations for tracking military movements in the Middle East.

Read Also:US Justice Department Blames 'Flawed' Work for Reflecting Pool Cracks

Probing Naval Vulnerabilities

Beyond mere tracking, the Iran-linked group used Claude to conduct technical research into the weaknesses of shipboard systems. The report highlights that the actors analyzed Cisco communications equipment, maritime satellite terminals, and various industrial control products to find potential points of failure. Furthermore, the AI was used to extract the names of US military personnel from publicly available photo captions, increasing the risk of targeted phishing attacks.

Why This Matters

BozokMedia analysis shows that this incident marks a critical shift in hybrid warfare. The ability of state actors to use Large Language Models (LLMs) to synthesize disparate data points into actionable military intelligence lowers the barrier to entry for sophisticated espionage. It proves that AI is no longer just for generating text; it is now an integral component of intelligence, surveillance, and reconnaissance (ISR) cycles.

"The convergence of OSINT and generative AI allows adversaries to achieve a level of situational awareness that previously required an entire intelligence agency."

Propaganda and Cyber Operations

The scope of the abuse extended far beyond naval intelligence. Anthropic identified separate accounts used to orchestrate massive propaganda campaigns. These accounts generated campaign plans and online personas, adapting Iranian government bulletins into multiple languages—including English, Spanish, Urdu, and Arabic—for distribution across X, Instagram, and TikTok.

Additionally, the report uncovered efforts to develop domestic surveillance tools, combining automatic license-plate recognition with mobile device interception. In the cyber realm, the actors used Claude to modify malware, making it more elusive to detection software and creating fake news outlets to spread disinformation.

Did You Know?: Anthropic was founded by former OpenAI executives with a specific focus on 'Constitutional AI,' a method designed to make AI systems safer and more transparent.

Frequently Asked Questions

Q1: Did the AI autonomously decide to help Iran?
No. The AI acted as a tool. The users provided the data and the prompts, and the AI performed the synthesis and organization tasks as requested, bypassing intended safety guardrails.

Q2: What actions has Anthropic taken since the discovery?
Anthropic has banned all identified accounts and shared critical threat intelligence with the US government and various technology partners to prevent future occurrences.