Anthropic PBC has reported that its artificial intelligence model, Claude, has been targeted for use in developing a wide range of potential military technologies, including kamikaze drone swarms, missile navigation systems, and biological weapons.
According to a threat intelligence report released by Anthropic on Thursday, users from Iran, Russia, China, and Yemen were among those identified misusing the company’s models over the past year. None of these incidents involved the company’s most advanced models, Fable and Mythos.
“As models become more powerful, the risks of their use will increase if AI developers and society’s defenders do not take steps to improve their safety,” Anthropic stated.
Attempts to use Claude for military development
Anthropic aims to prevent its technology from being used for harmful purposes, including the development of biological and conventional weapons. This stance has led to direct friction with the Pentagon, which suspended work with the company after it refused to allow Claude to be used for mass surveillance or autonomous weapons systems.
The Anthropic report details instances where users attempted to bypass Claude’s safety guardrails or mask their locations. Notably, users in Russia—whom Anthropic does not believe are linked to the Russian state—attempted to create software for an autonomous first-person-view kamikaze drone swarm. They used combat footage from Ukraine to train the system.
In northern Yemen, currently under Houthi control, local users utilized Claude to develop software for several types of missiles. However, Anthropic stated it has no evidence that these users were able to build and deploy a functional system.
The report also includes five examples of Claude being used in ways that could facilitate the development of biological weapons. This includes a case where virologists linked to state entities attempted to create an enhanced version of the mosquito-borne chikungunya virus.
This attempt was discovered following a query to Claude for help writing a grant application to fund research at a military institute in an unnamed country where Anthropic does not provide services. The company emphasized that all five biological cases described in the report were ambiguous and could have involved research with entirely legitimate scientific goals.
AI used for surveillance and cyberattacks
State-linked actors in China and Iran have attempted to use Claude to develop surveillance systems. In one Chinese operation, Claude was used to “track, profile, and recruit Uyghurs” within the Syrian military.
An Iranian unit used Claude to create surveillance software disguised as a prayer-time tracking tool.
In another instance, Anthropic discovered that a consultant, allegedly working for a state intelligence agency in Mali, used Claude to build a monitoring system for approximately 25 million mobile phones. The system could generate intelligence dossiers on any specified phone number without a court order.
Anthropic blocked the consultant’s account, but the Claude-based surveillance platform had already been deployed locally and was beyond the company’s full control. The government of Mali did not respond to a request for comment.
The report also describes the use of AI by hackers—ranging from suspected state intelligence operatives in Russia and China to hacktivists and cybercriminals—to accelerate and scale the theft of sensitive data.
Specifically, the hacking group Midnight Blizzard, which Microsoft links to Russian intelligence services, used AI to automate many stages of malware creation and to prepare phishing emails targeting military intelligence and defense organizations.
Another group, ShinyHunters, which Anthropic describes as a cyber-extortion gang, used AI to search for leaked credentials and subsequently used them to steal data from victimized organizations.
Anthropic emphasized that these cases do not indicate the emergence of new, dangerous vulnerabilities in critical systems. The company described them as examples of the most notable and novel activity involving the misuse of its technology that it has identified to date.
Source: Bloomberg
