Anthropic Report Details Iranian Propaganda Operations and Yemen Missile Development Using Claude

Anthropic said Friday it had identified and disrupted multiple operations linked to Iranian state institutions that used its Claude AI models for propaganda, surveillance and military-related activities. Separately, the company uncovered a weapons-development cell in northern Yemen that used Claude to develop guidance software for missiles and rockets.

The findings were detailed in Anthropic’s latest threat intelligence report, which covers malicious activity detected between December 2025 and August 2026. The company said it identified and disrupted three Iranian state-aligned influence operations connected to the Islamic Culture and Communications Organization (ICCO), the Islamic Propaganda Office of Khorasan Razavi and the Islamic Propaganda Organization’s Bina Cultural Observatory.

According to Anthropic, the Iranian operations were designed as part of what their operators described as a “soft war” or “cognitive warfare” campaign intended to shape public opinion inside and outside Iran. The operators used Claude to develop campaign plans, doctrine manuals, online personas, target databases and ministerial planning documents.

The report says the operations were explicitly tied to Iran’s state doctrine of “Jihad al-Tabyin,” or “explanatory jihad,” under which Iranian institutions produce propaganda as both a religious and strategic duty.

One network reportedly operated a multi-province content factory known as “Manjanegh,” or “Catapult,” which used dozens of activists to repackage Iranian security services’ public reporting under different personas. Anthropic said the network amplified the material through paid campaigns across more than 100 Iranian platform channels, including channels affiliated with the Islamic Revolutionary Guard Corps.

The company also linked a director-level official at the Bina Cultural Observatory to the operation. Anthropic said the actor generated messaging in the official voice of an IRGC spokesperson and, during the 2026 U.S.-Israel-Iran war, attributed false claims to Western research institutions including CSIS, Brookings and RAND in an apparent effort to make Iranian state-backed narratives appear more credible.

Claude was used to transform Iranian government intelligence bulletins into content in Farsi, Arabic, Urdu, Malay, Spanish and English, with the operators planning to expand the operation to 20 languages. The system was also used to disguise the origin of propaganda by making material appear to have been written by foreign journalists, independent news organizations or ordinary citizens.

Anthropic said the Iranian actors attempted to conceal their identities because access to Claude from inside Iran is blocked. They reportedly used VPNs and foreign phone numbers to register accounts, but repeatedly disclosed their locations, institutions and roles during their interactions with the model.

The report also details separate Iran-linked military and surveillance activity. One Iran-nexus actor used Claude to collect and analyze publicly available information on U.S. naval forces in the region and compile targeting handbooks. The material included information on U.S. personnel, ship and aircraft identifiers, satellite imagery and publicly exposed websites tracking naval movements. The actor also used Claude to research vulnerabilities affecting maritime communications and industrial systems.

Another Iran-nexus operation used Claude to build an automated open-source intelligence system capable of profiling hundreds of individuals in Israel and the Jewish diaspora, while a separate workstream involved modifying malware and developing techniques intended to make it harder to detect.

Anthropic additionally identified Iranian actors using Claude to support domestic surveillance systems, including software combining automatic license-plate recognition with mobile-device identifier interception.

Separately, Anthropic identified a weapons-development cell based in northern Yemen that used Claude Code to develop guidance, navigation and control software for guided weapons.

The cell was working on three weapons programs: a guided rocket using a phone-class flight computer and terminal homing guidance, a multi-stage ballistic missile with a stated range exceeding 2,000 kilometers, and a multi-variant missile program known as the “R2000” set, including a hypersonic-glide-vehicle variant.

Anthropic said the operators used Claude Code in place of human software engineers to develop guidance, navigation and control software. Multiple Claude instances were assigned different roles, including coding, research and review, effectively allowing the operators to use the AI as a small engineering team.

The company said the actors test-fired a guided rocket in Yemen. The test appeared to fail, after which the operators returned to Claude within hours to analyze the failure. Anthropic said it had no evidence that the group ultimately succeeded in fielding an operational weapon.

The report says the actors attempted to circumvent Claude's safeguards by concealing the purpose of their work and dividing their activities across multiple sessions so that no single conversation exposed the full weapons-development program.

Anthropic ultimately banned the accounts associated with the Yemen operation and shared threat intelligence with public- and private-sector partners. However, the company said the actors had already built an offline simulation toolkit capable of operating without Claude or other engineering platforms.

Stay sharp with the stories, insights, and developments shaping Lebanon, delivered straight to your inbox weekly.

By signing up, you acknowledge and agree to our Privacy Policy and Terms of Service . You may unsubscribe at any time by following the directions at the bottom of the email.