Illustration: a dossier collage of a typewriter, drafted press releases and an annotated speech transcript.

Three Iranian propaganda offices used Claude as a staff shop to plan soft-war personas and paste IRGC lines onto Western think-tank letterhead.

Illustration, not an archive object

Filed
September 17, 2026, 02:05 UTC
Actor
Islamic Culture and Communications Organization, Islamic Propaganda Office of Khorasan Razavi, and Islamic Propaganda Organization Bina Cultural Observatory, using Anthropic's Claude under GTG-34001
Target
Iranian domestic audiences and foreign readers asked to treat IRGC-aligned copy as independent commentary or as research from named Western institutes
Source
Unidentified Source blurred: contractors, cutouts, buried disclosure
Confidence
high Multiple independent open sources, low disagreement about the facts
Techniques
Synthetic mediaNarrative launderingTestimonialManufactured consensusTransfer
Where
Iran, United States
Channels
  • Anthropic Claude accounts banned under GTG-34001
  • Islamic Culture and Communications Organization ministerial deliverables and cultural-attaché network
  • Mashhad seminary cognitive-warfare command room and Manjanegh (Catapult) content factory
  • Paid amplification across more than 100 Iranian platform channels, including IRGC-tied handles
  • False attribution to CSIS, Brookings and RAND during the 2026 U.S.-Israel-Iran war campaign
  • Cross-posting plans for Eitaa, Bale, Rubika, X, Instagram, Telegram, TikTok and YouTube

Summary

On 10 September 2026 Anthropic published Detecting and countering misuse of AI, covering activity disrupted between December 2025 and August 2026. Case GTG-34001 names three Iranian state-aligned accounts that used Claude to stand up influence campaigns. Anthropic tied the operators to the Islamic Culture and Communications Organization under the Ministry of Culture and Islamic Guidance, the Islamic Propaganda Office of Khorasan Razavi running a cognitive-warfare room out of a Mashhad seminary, and the Islamic Propaganda Organization's Bina Cultural Observatory. The company said the operators described a non-military plan to shape public opinion at home and abroad, asked the model for campaign plans, doctrine manuals, persona systems, target databases and ministerial paperwork, and in one war-week thread attributed false claims to CSIS, Brookings and RAND. RFE/RL carried the same named institutions on 11 September. Score the named shops, the product list, and the think-tank attribution as high. Score a finding that any given viral clip this week was generated by these three accounts as not in this file.

Analysis

The last two briefs in this vault covered a Kremlin-linked Cameo-and-CNN-bug midterm drop and a Pentagon SGE influencer payroll. This cycle rotates to Iranian state propaganda institutions that treated a U.S. commercial model as a program office.

Primary layers agree on the shops and the product list. Anthropic's 10 September 2026 threat report, case GTG-34001, says it identified and removed three Iranian state-aligned accounts using Claude to set up influence campaigns. The operators were planning content for what they called a soft-war or cognitive-warfare program; in their own words, a non-military plan to shape public opinion at home and abroad. Each operation, the company wrote, was run by an actor working within or on behalf of a named Iranian state propaganda institution: the Islamic Culture and Communications Organization under the Ministry of Culture and Islamic Guidance; the Islamic Propaganda Office of Khorasan Razavi, running a cognitive-warfare command room out of a Mashhad seminary and distributing IRGC-aligned content; and the Islamic Propaganda Organization's Bina Cultural Observatory. The report lists the model work as campaign plans, doctrine manuals, persona systems, target databases and ministerial planning documentation. It adds that the network produced ICCO-branded ministerial deliverables for a nine-part international influence portfolio and complete organizational plans for the funeral of Iran's Supreme Leader; that Mashhad leaders operated Manjanegh, a multi-province factory that repackaged security-service public reporting under personas with no visible security-service link, then bought amplification on more than 100 Iranian channels including IRGC-tied ones; and that a Bina director-level official generated messaging in an IRGC spokesperson's official voice and, in a campaign around the 2026 U.S.-Israel-Iran war, attributed false claims to CSIS, Brookings and RAND. RFE/RL, 11 September, restated the three institutions and the program-office quote. Score those dated texts as high. Score a signed tasking cable from a named IRGC desk, and any claim that every anti-Israel post this month came from these three accounts, as not shown.

Mechanism is synthetic-media first. The model is not a witness. It is a staff shop that writes doctrine, personas and calendars a ministry would otherwise have to hire. Narrative-laundering is the architecture. Security-service copy is stripped of the uniform, given a civilian face, and sent through cultural attaches whose job, in one reconstructed line, is not to be the narrator but the director. Testimonial is the war-week cut. A sentence that would be discarded as IRGC communique is offered as CSIS, Brookings or RAND. Transfer pastes the prestige of those letterheads onto a seminary command room. Manufactured-consensus is the paid layer: more than 100 channels, one factory, so a thin line looks independently discovered.

This is propaganda where origin is three named state shops, institutional form is a commercial chatbot plus cultural-office letterhead, and the design goal is to move identification so a reader treats a persona or a think-tank name as the warrant. Gray does not cancel the method. It is not a finding that every Iranian official who uses a writing tool is running an influence net, that Anthropic invented the institutions, that MEK impersonation in a separate case is the same operation, or that banning the accounts ended the doctrine. A same-method test applies in reverse. A Western ministry that used a frontier model to draft persona kits, buy 100 friendly channels, and paste its war copy onto named Iranian research institutes would be this brief with the nouns swapped.

Literacy counter: keep three objects apart. One: Anthropic banned three accounts and named three offices. That is a platform record. Two: a real argument about electricity, hostages, or a strike. That can be checked without a persona. Three: the instruction that a Brookings or RAND sentence, or an ordinary-looking channel, is how you know the line is independent. If the payload is two or three, the model run is doing method work. If the payload is one, open the GTG-34001 section, the ICCO masthead, and the think-tank page that did not publish the claim on the same screen.

Tells

Evidence

  1. Anthropic threat report, 10 September 2026: GTG-34001 naming ICCO, Mashhad Propaganda Office, and Bina Observatory
  2. Anthropic PDF, Detecting and countering misuse of AI, September 2026
  3. RFE/RL, Frud Bezhan, 11 September 2026: same three institutions, program-office quote, MEK companion case
  4. Unite.AI, 10 September 2026: secondary reconstruction of influence-section cases and GTG designators