Anthropic says Iran, Russia used Claude for weapons research, including missile systems and kamikaze drone projects

Anthropic's artificial intelligence model Claude has been misused in attempts to develop military applications including kamikaze drone swarms and missile navigation systems, as well as research linked to potential biological weapons, according to a report by the company's threat intelligence team.

The 154-page report, cited by Bloomberg, said users from countries including Iran, Russia, China and Yemen were among those flagged for misusing Anthropic's frontier models over the past year. None of the cases involved the company's two most advanced models, Fable and Mythos.

"As models become increasingly capable, their risks will increase, unless AI developers and society's defenders act to make them safer," Anthropic said in the report.

According to the report, users attempted to evade Claude's safeguards or conceal their locations while using the AI for potentially harmful purposes.

Users in Russia, whom Anthropic determined were not linked to the state, tried to use Claude to build software for an autonomous first-person-view kamikaze drone swarm trained on Ukrainian combat footage.

Actors based in Houthi-controlled northern Yemen used Claude to build software for several types of missiles. Anthropic said it had no evidence that the actors succeeded in fielding an operational device, but said they had test-fired a guided rocket. Hours later, the actors returned to Claude to determine why the rocket had failed.

The report also detailed five cases involving the potential development of biological weapons. These included state-sponsored virologists seeking to create an enhanced version of chikungunya, a debilitating mosquito-borne virus.

Anthropic said the biological cases were "ambiguous" and involved research that could have benign scientific purposes. In one case, the effort was identified through a request for Claude to help write a grant application for research at a military institute in an unnamed country where Anthropic does not provide service.

Surveillance efforts

State-linked actors in China and Iran also allegedly used Claude for surveillance-related activities.

According to Anthropic, one China-linked operation used Claude to track, profile and recruit Uyghurs in the Syrian Army. An Iranian unit used the AI to build surveillance software disguised as a tool for tracking prayer times.

In another case, Anthropic identified a consultant likely working for a state intelligence agency in Mali who used Claude to build a system designed to monitor about 25 million mobile phones in the West African country.

The system could generate intelligence dossiers on specified phone numbers without a court order, Anthropic said. The company banned the consultant's account, but said the Claude-powered surveillance platform had already been deployed locally and was outside its full control.

AI used to accelerate hacking

Anthropic's report also highlighted the use of Claude by hackers, including suspected state-sponsored intelligence groups in Russia and China, hacktivists and cybercriminals.

The report said one hacking group, identified by Microsoft as Midnight Blizzard and linked to Russia's intelligence services, increased the speed of its operations by automating them using AI. The group used Anthropic's software to automate parts of malware development and help craft phishing emails targeting military intelligence and defence organisations.

Anthropic also said the ShinyHunters cyber-extortion gang used AI to scan for leaked security credentials and use them to steal data from affected organisations.

However, the report did not identify any novel or dangerous new vulnerabilities in critical systems.

Anthropic-Pentagon conflict

The report comes as Anthropic faces heightened scrutiny over how its AI systems should be used by governments and militaries.

The company's position has put it at odds with the Pentagon, which suspended work with Anthropic after the company refused to allow Claude to be used for mass surveillance or autonomous weapons systems.

Anthropic said the cases described in its report were not typical examples of misuse but represented some of the most notable and novel threat activity identified by its team. The report also covered other forms of misuse, including scams and hacking tools.

The findings come amid renewed debate over the longer-term risks of increasingly capable AI systems. Researcher Jacob Coxon resigned from Anthropic on Tuesday, saying the company and OpenAI were "gambling with our lives" in their pursuit of increasingly capable AI.