Iran-linked group used Claude to try targeting U.S. warships, Anthropic says

Report also details Yemen missile work, Uyghur surveillance, Russian drones

Anthropic said in a 154-page report this week that an Iran-linked group used the company’s Claude artificial-intelligence model to compile targeting information on U.S. Navy warships operating in the Middle East, and that the company disrupted the effort.

According to the report, the Iranian-linked group gathered data from ship and aircraft transponders, military photographs and commercial satellite photos to track American warships and look for vulnerabilities in their ship-born communications. The finding was one of several threats outlined in the 154-page report, which Anthropic described as a playbook of how U.S. adversaries are using AI in novel ways that pose serious threats to national security — including developing new missiles and aggressive disinformation campaigns, spying on dissidents, and even pursuing the potential development of biological weapons.

Anthropic said it identified and disrupted efforts to use its AI models to develop biological weapons, though the company did not specify which countries or foreign governments were linked to those efforts. “Without the correct safeguards, such capabilities could have catastrophic consequences,” the report said.

The same report documented other adversaries’ use of Claude:

A group based in northern Yemen used the model to develop guidance software for high-end weapons including a multistage ballistic missile, the report said. Yemen is home to the Houthi rebel group, which has emerged as one of the strongest Iran-aligned groups in the Middle East and recently made gains to consolidate control over the strategic Bab al-Mandeb waterway that acts as a chokepoint for energy shipments through the Red Sea. “Our safeguards blocked many of their requests, but not all of them,” the Anthropic report said of the Yemeni weapons-development efforts. The company said it banned the accounts associated with this effort, shared the information with government partners and developed new detection protocols to reduce the future risk of misuse.

A Chinese-government-aligned operation used Claude to track and surveil Uyghurs — an ethnic minority primarily based in western China — including those who have joined the newly formed Syrian Army, according to the report. The same operation ran mass reporting and bot-network amplification campaigns against journalists from the Uyghur diaspora. The U.S. declared Chinese government repression of Uyghurs a genocide in 2021.

Another set of accounts linked to the Chinese government used Claude to develop dossiers on religious leaders across Asia that Beijing views with suspicion. Targets identified in the report included senior Catholic cardinals across Asia, the leadership of the Presbyterian Church in Taiwan, members of Tibetan Buddhist civil society, and the administration in exile.

Russia-based actors attempted to use Claude to build autonomous killer drone swarms using data from Ukrainian battlefields and develop disinformation campaigns by distributing AI-generated Russian-aligned state propaganda in resource-rich countries in central Africa, including the Central African Republic and Democratic Republic of Congo, according to the report.

Anthropic’s rules don’t allow Claude to be used to build weapons, conduct surveillance or carry out violence. But people often sought to obscure their intentions and persistently tried new approaches when Claude rejected their requests, the report said.

The report was published the same week that a top researcher at Anthropic quit over fears that AI companies were developing systems they could not control, the Wall Street Journal reported. Other Anthropic employees have warned that AI could wipe out humanity if not properly contained.

In a social-media post quoted by the Journal, Evan Hubinger, another top scientist at Anthropic, wrote: “We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”

Anthropic said it hoped the report’s findings would help other developers recognize similar patterns on their platforms, give governments and civil society a clearer view of how emerging threats take shape, and strengthen collective defenses.