Anthropic Disrupts Iranian State Propaganda Operations Built on Claude

Anthropic has published its latest threat intelligence report detailing AI misuse it detected and disrupted between December 2025 and August 2026, and one case in particular lands squarely on the conflict this site has tracked all week: three separate Iranian state-aligned accounts that used Claude to build out what their own operators described as a "soft war" or "cognitive warfare" programme aimed at shaping public opinion at home and abroad.

Anthropic traced each operation to a named Iranian institution. One ran through the Islamic Culture and Communications Organization, which sits under Iran's Ministry of Culture and Islamic Guidance. A second operated a "cognitive warfare command room" out of a seminary in Mashhad, run by the Islamic Propaganda Office of Khorasan Razavi and distributing content aligned with IRGC narratives. A third traced to a director-level official at the Islamic Propaganda Organization's Bina Cultural Observatory. Across all three, the actors used Claude to generate campaign doctrine manuals, persona systems, target databases and ministerial planning documents, the kind of output that would otherwise require a fully staffed program office to produce.

The operations were explicit about their strategic framing. Anthropic found the actors repeatedly invoking Iran's state doctrine of "Jihad al-Tabyin," or explanatory jihad, under which Iranian institutions treat propaganda production as both a religious and strategic duty, language that appeared directly inside the accounts' planning documents and sessions. One network produced a complete nine-part international influence portfolio carrying official institutional branding, alongside organisational plans for the funeral of Iran's Supreme Leader.

Attribution laundering was a consistent thread across all three cases. The Mashhad-based network, which Anthropic linked to publicly identifiable leaders operating a multi-province content operation known as "Manjanegh" (Catapult), used dozens of activists to repackage Iranian security services' public reporting under invented personas with no visible link back to the state. The Bina Observatory-linked actor went further, generating messaging in the voice of an actual IRGC spokesperson across multiple sessions, and during the 2026 US-Israel-Iran war specifically, attributed fabricated claims to Western research institutions, naming CSIS, Brookings and RAND, to make Iranian-originated messaging read as independently corroborated by credible outside sources.

The operations also targeted specific groups beyond general public messaging. Anthropic's investigation found aggressive counter-narrative content aimed at the Bahá'í, a persecuted religious minority in Iran, alongside target databases naming international officials and Iranian opposition figures. Notably, Claude reportedly refused the operation's most aggressive requests, including one instance where it declined to name real individuals as militants in a way that could draw security action against them, forcing the operators to fall back on anonymous-source framing instead.

Anthropic assessed the overall reach of the Iranian operations at Category Three on the Breakout Scale, a six-tier framework used across the industry to measure how far an influence operation's content actually spreads, meaning content was observed across multiple platforms with some pickup by IRGC-aligned distribution channels, short of the wider cross-platform breakout that would put it at the top of the scale.

This finding sits directly alongside the broader pattern of AI-accelerated state activity this site has already documented through the current conflict. Iranian state-linked hackers have separately been observed using fake recruiter personas on LinkedIn to deliver new cross-platform malware specifically targeting critical sectors across the Middle East, and the UAE's own cybersecurity chief has publicly called for an international "Cyber Geneva Convention" specifically citing AI's compression of the time available to detect and respond to combined cyber and information threats. Read together, the propaganda apparatus Anthropic disrupted is not a separate front from the cyberattacks and drone strikes already documented in this conflict, it is the same state applying the same AI-driven speed advantage to its information operations that other Iranian-linked actors have applied to malware development and network intrusion.

For GCC governments and enterprises, the practical takeaway is that Iranian information operations are not confined to fringe accounts or obviously fabricated content. These operations were built to be indistinguishable from independent local commentary, laundered through fabricated institutional attribution, and specifically engineered to evade the kind of surface-level detection that communications and public-affairs teams typically rely on. Anthropic's disruption came from detecting the activity during content production, before it reached a public platform, underscoring that upstream detection at the AI-platform level, not just downstream content moderation, is now a meaningful part of the defensive picture against state-directed influence campaigns.