Anthropic says Iran-linked actors used its AI for propaganda and spying

WASHINGTON (MNTV) — The U.S. artificial-intelligence company Anthropic says Iranian state-aligned and Iran-linked accounts used its Claude AI models in a series of covert operations — spanning propaganda, domestic surveillance, and the targeting of U.S. naval forces — that it detected and shut down over the past nine months.

The disclosures appear in Anthropic’s latest threat-intelligence report, published Sept. 10, which catalogs misuse of Claude between December 2025 and August 2026 across seven areas including cyber operations, influence campaigns, and surveillance.

Because the report comes from the company whose product was involved, its findings reflect Anthropic’s own investigation and attribution; the firm said it banned the accounts, tightened its safeguards, and shared intelligence with authorities where appropriate, while noting that disrupting the accounts did not necessarily end the underlying activity.

On influence operations, Anthropic said it removed three Iran state-aligned accounts it linked to government propaganda bodies — the Islamic Culture and Communications Organization under Iran’s culture ministry, an Islamic-propaganda office running what the report called a “cognitive warfare command room” out of a Mashhad seminary, and the Bina Cultural Observatory.

The operators described their work as a “soft war” to shape opinion at home and abroad, tying it to Iran’s state doctrine of “Jihad al-Tabyin,” or explanatory jihad, under which propaganda is treated as a religious and strategic duty, the report said.

They used Claude to build campaign plans, doctrine manuals, fake-persona systems, and target databases, engineering content so that state narratives appeared as independent voices — one operator describing the role as “not to be the narrator, but the director.”

One account generated messaging in the official voice of a spokesperson for the Islamic Revolutionary Guard Corps (IRGC), Iran’s elite military branch, and during the 2026 U.S.-Israeli war on Iran the network falsely attributed claims to Western research institutions — the Center for Strategic and International Studies, the Brookings Institution, and RAND — to make its messaging appear more credible, according to the report.

Anthropic said the network also deployed “aggressive counter-narrative content” targeting the Bahá’í, a religious minority long persecuted in Iran, and built databases naming international officials and Iranian opposition figures.

Anthropic said other Iran-linked accounts turned to surveillance and military targeting.

One used Claude to gather and analyze publicly available information to compile “targeting handbooks” and track the positions of U.S. naval forces in the region, it said.

Another built a malicious Firefox browser extension designed to harvest users’ identities from social-media platforms, and developed a method to identify individuals through their social-media accounts — capabilities that rights groups warn are typically turned against dissidents, minorities, and government critics.

The company’s head of threat intelligence, Jacob Klein, said AI was not changing whom governments target but was making surveillance “cheaper and more efficient,” letting a single official or contractor automate work that once required teams of analysts.

The claims about Iranian activity, drawn from Anthropic’s own monitoring, could not be independently verified, and Iran has not publicly responded. The report adds to a broader picture, sketched across the AI industry, of governments increasingly folding commercial AI tools into the machinery of propaganda and state surveillance.