Anthropic Says Iranian Groups Used Claude AI to Run Propaganda Campaigns

By : Alia Kashmiri ( Editor)

Anthropic has said three Iranian state-linked influence operations used its Claude AI system to help organise propaganda campaigns and spread government-backed messages online.

The company said the operations were linked to three Iranian propaganda institutions: the Islamic Culture and Communications Organization, the Islamic Propaganda Office of Khorasan Razavi and the Bina Cultural Observatory of the Islamic Propaganda Organization. Click Here To Follow Our WhatsApp Channel

According to Anthropic, the groups used Claude to plan campaigns, prepare operational documents, create online personas and develop strategies for reaching audiences in Iran and other countries.

The company said the operations were part of programmes described by those involved as a “soft war” or “cognitive warfare”, aimed at influencing public opinion both inside Iran and abroad.

Claude used to organise propaganda campaigns

Anthropic said Claude was used as a central tool for organising the campaigns. The groups reportedly created operation manuals, lists of target audiences, online personas, monitoring systems and plans for spreading content.

They also used the AI system to adapt official government information into content in several languages, including Farsi, Arabic, Urdu, Malay, Spanish and English. The groups reportedly planned to expand their operations to as many as 20 languages.

A major focus was reportedly placed on hiding the government links behind the campaigns. According to Anthropic, some actors created posts designed to appear as though they had been written by independent journalists, foreign writers or ordinary social media users.

The content was shared on Iranian platforms such as Eitaa, Bale and Rubika, as well as international services including X, Instagram, Telegram, TikTok and YouTube.

Groups tried to hide their identities

Anthropic said the operators took steps to conceal their locations and connections because Claude is not available for use inside Iran.

The company said they used VPNs and foreign phone numbers to create and verify accounts. However, some of the users later revealed details about their organisations and roles while interacting with Claude.

Anthropic said documents carrying institutional branding and information available from open sources also helped investigators connect the accounts to the Iranian organisations.

The company subsequently banned the accounts linked to all three operations and shared information about the activity with industry and research partners.

AI-generated content targeted foreign audiences

One of the operations reportedly produced official documents outlining an international influence programme. Anthropic also said the group prepared organisational plans connected to the funeral of Iran’s Supreme Leader Ali Khamenei, who was killed during the early stages of the war.

Another operation based in Mashhad reportedly ran a multi-province content network called “Manjanegh”, meaning “Catapult”.

According to Anthropic, the network involved dozens of activists who repackaged publicly available reports from Iranian security services under different identities. The material was then promoted through paid campaigns across more than 100 Iranian channels, including outlets linked to the Islamic Revolutionary Guard Corps (IRGC).

Anthropic also identified an operation linked to the Bina Cultural Observatory that generated messages using the official voice of an IRGC spokesperson.

During a campaign linked to the 2026 US-Israel-Iran war, the network allegedly attributed false claims to Western research organisations, including CSIS, Brookings and RAND. The aim, Anthropic said, was to make government-backed messages appear more credible.

The report said the operations also targeted international officials, Iranian opposition figures and the Baha’i community.

Anthropic said the use of AI allowed the groups to produce and distribute large amounts of content while making some state-backed messages appear to come from independent sources.