A recent report by artificial intelligence developer Anthropic has highlighted the use of AI technology by Iranian-linked institutions for surveillance and propaganda purposes. The investigation traced influence operations involving the company’s Claude models to three organizations tied to the Iranian government: the Islamic Culture and Communications Organisation, the Islamic Propaganda Office of Khorasan Razavi, and the Bina Cultural Observatory.

According to the report, despite AI tools like Claude not being officially available in Iran, Iranian actors employed the technology to design misinformation campaigns, compile detailed target databases, draft official government documents, and even develop plans related to Iran’s supreme leader. These activities demonstrate a sophisticated integration of commercial AI into state propaganda and intelligence operations.

The report outlines how Iranian propaganda units framed their efforts as a form of religious and strategic duty, invoking the concept of Jihad al-Taybin, or "explanatory jihad." AI was used to camouflage government narratives as independent commentary and attribute false information to Western research bodies. This approach aids in amplifying Tehran’s messages while obscuring their origin, facilitating more effective influence across multiple languages and audiences.

Anthropic also revealed development of surveillance tools by Iran-linked groups, including an identity-harvesting browser extension disguised as a prayer-time utility. Such tools collect personal data from social media accounts, enabling authorities to map individuals’ relationships and potentially leverage threats against relatives to control dissidents. These methods threaten trust within communities, complicating efforts by activists and ordinary citizens to organize or speak freely without fear of reprisal.

The scope of Iranian AI misuse documented by Anthropic extends beyond dissidents inside Iran. The report identifies operations targeting hundreds of individuals and organizations in Israel and the Jewish diaspora, as well as activities resembling military reconnaissance against U.S. naval forces. This breadth highlights an extensive use of AI-enabled tools to support Tehran’s strategic interests and surveillance capabilities.

These findings carry particular significance for Australia, home to sizable Iranian and Jewish communities. Australian security agencies have previously raised concerns about Iranian surveillance and directed attacks on religious and community centers, including incidents in Sydney and Melbourne in 2025. While AI involvement in these events has not been confirmed, the report underscores the evolving risks posed by foreign state actors using AI-assisted methods on Australian soil.

Anthropic acknowledged that some surveillance-related applications were made possible by gaps in its AI safeguards. The company has since banned dozens of accounts linked to Iranian state actors but noted that the deletion of accounts does not erase collected data or disable tools already deployed. Experts emphasize the need for technology providers and democratic governments to enhance protections for individuals exposed to foreign AI-driven surveillance, including confidential reporting mechanisms and targeted support.

For those facing political repression, AI safety is intertwined with their ability to communicate, organize, and maintain trust without endangering themselves or their families. The report stresses that ensuring AI remains under human control must consider whose interests it serves and how vulnerable populations can be shielded when such technologies are employed by authoritarian regimes.