New york: Anthropic has reported a significant disruption in attempts by Iran, China, and other nations to utilize its artificial intelligence models for spreading state propaganda and conducting surveillance on ethnic minorities and dissidents.
According to Radio Free Europe Radio Liberty, the AI company's latest threat intelligence report, released on September 11, outlines incidents spanning from December 2025 to August. It reveals that criminal entities, state institutions, and scientists, particularly in Russia and China, have been employing AI models like Claude and Gemini for designing missiles, bombs, and even creating deadly pathogens.
Anthropic highlighted that several Iranian state institutions have been using its AI models to influence public opinion both domestically and internationally. Key among these entities are the Islamic Culture and Communications Organization, the Islamic Propaganda Office, and the Islamic Propaganda Organization. These organizations are accused of executing "cognitive warfare" campaigns and promoting narratives developed by the Islamic Revolutionary Guards Corps, a powerful faction of Iran's armed forces.
The report states, "In each case, the operation relied on Claude to build campaign plans, doctrine manuals, persona systems, target databases, and ministerial planning documentation." This usage of AI allowed the institutions to create complex organizational assets that would typically require a fully staffed program office.
One notable instance involves the Islamic Culture and Communications Organization using Claude to complete organizational plans for the state funeral of the late Supreme Leader Ayatollah Ali Khamenei, held from July 3-9.
In a separate revelation, Anthropic disclosed its disruption of an "influence operation" by the Mujahedin-e Khalq (MEK), an exiled opposition group classified as a terrorist organization by Tehran. This operation utilized a shared AI agent to impersonate individuals and recruit within Iran. The operation involved scraping over 500 social media channels to compile detailed profiles, categorizing individuals by city, age, occupation, political alignment, and arrest history to tailor recruitment messages.
Additionally, Anthropic thwarted surveillance operations by state actors, including Iran, that employed Claude to facilitate these activities. One such operation involved Iran-based actors using Claude to develop a malicious Firefox extension for harvesting user identities from social media. Another case saw Iranian actors analyzing social media posts to monitor 39 opposition accounts.
"The operators were state-aligned organizations targeting the same diaspora and dissident communities historically pursued by these regimes," Anthropic reported.