Iran, China and a Yemeni Missile Cell All Tried to Weaponize Claude. Anthropic Caught Them.

Anthropic’s Threat Intelligence team spent eight months tracking foreign operatives, criminal syndicates, and at least one weapons cell as they tried to turn the company’s Claude models into tools for war, espionage, and repression.

The revelation comes via a new report the AI lab published this week. The report details the activity Anthropic says it identified and shut down between December 2025 and August 2026.

It’s the company’s fourth such disclosure since March 2025, and it lands the same week Washington erupted over separate warnings about long-term AI risk. Anthropic’s own framing is less about the technology’s distant future and more about what’s already happening: the same capabilities that make Claude useful for coding, research, and customer service also lower the bar for hostile actors who used to need whole teams of specialists to pull off similar operations.

An Iran-Linked Network Went After U.S. Naval Movements

One of the operations Anthropic disrupted involved an Iran-linked group that used Claude to help build targeting handbooks tracking the positions of U.S. ships, personnel, aircraft, and identifying markers, along with satellite imagery and open websites that expose naval movements. Separate Iran-linked efforts used the model for propaganda production, domestic surveillance, and the targeting of opposition figures and minorities inside Iran.

Anthropic’s broader report also documents three Iranian state-aligned propaganda accounts, tied to the Ministry of Culture and Islamic Guidance’s Islamic Culture and Communications Organization, an Islamic Propaganda Office command room in Mashhad, and the Bina Cultural Observatory, that used Claude to build campaign doctrine, persona systems, and target databases in support of what the actors themselves called a “cognitive warfare” program.

A Weapons Cell in Yemen, a Hunt for Uyghurs in China, and a Surveillance Network

A team in northern Yemen relied on Claude Code, Anthropic’s coding tool, to develop guidance software for rockets and missiles, running separate instances of the model to write code, conduct research, and check each other’s work. After one guided-rocket test apparently failed, the team returned to Claude within hours to help diagnose what went wrong.

Anthropic also says a China-linked operation combed through more than 100 WhatsApp groups and dozens of Telegram channels to identify Uyghurs in Syria who could be pressured or paid to inform on armed Uyghur factions. Claude then helped a non-Arabic-speaking operator conduct outreach in Syrian Arabic, translating replies and coaching efforts that exploited targets’ money troubles, family separation and relatives still living in Xinjiang.

And a consultant in Mali used Claude as the primary engineering force behind a nationwide system capable of collecting call records, text messages, and voice traffic across the country, the report states. The system could reportedly identify individuals by voice across different SIM cards, flag VPN users, and generate intelligence dossiers on any phone number without a warrant.

Claude Refused a Bioweapons Request. The Researchers Went Elsewhere.

In a case Anthropic frames as evidence its safeguards work, researchers operating under a state-backed grant tried to use Claude to alter chikungunya, a mosquito-borne virus, to make it spread more easily or evade immune defenses, work intended for a military research institute. Claude blocked the most sensitive requests. The researchers then routed the work to a competing AI model with weaker safeguards, a detail Anthropic includes as a reminder that one company’s safety rules only go so far when rivals don’t apply the same standard.

Anthropic Frames Itself as an Early-Warning System, Not Just a Target

Company researchers argue the same access that lets bad actors misuse Claude also gives Anthropic an unusually early window into emerging threats. They point to a separate discovery, Russian-designed drones built to select human targets without human sign-off, as an example of the intelligence value the company says it can offer governments and industry partners. The report reads less like an indictment of the technology and more like a company using its own product to police misuse of it, then publishing what it found instead of sitting on it.

Washington Reacts, But Not With One Voice

The report landed in the middle of a separate, broader fight over AI safety that’s consumed Capitol Hill this week. A bipartisan group of House members is pushing Speaker Mike Johnson to cancel recess until Congress acts, while Sen. Bernie Sanders, one of Congress’s most vocal AI critics, has called for banning superintelligent systems outright. 

President Trump has taken the opposite tack, playing down extinction-level fears and arguing the bigger danger is ceding the AI race to China, a framing that lines up with the administration’s broader push to accelerate American AI development rather than throttle it.

Editor’s Note: Thanks to President Trump and his administration’s bold leadership, we are once again respected on the world stage, and our enemies are being put on notice.

Help us continue to report on the administration’s peace through strength foreign policy and its successes. Join RedState VIP and use promo code FIGHT to receive 60% off your membership.