China’s People’s Liberation Army is using one of America’s most advanced artificial intelligence agents to develop weapons, conduct cyber espionage operations and impose draconian surveillance controls, according to an alarming threat intelligence report by Anthropic.
Russia, Iran and Yemen terrorists were also detected by the AI company and its agent Claude conducting weapons work and government information and intelligence operations against the United States. That included Tehran’s use of American AI to target U.S. naval forces in the Middle East, the 154-page report states.
An unspecified Iranian threat actor used Claude to gather and analyze information on Navy forces in the region to “develop targeting recommendations,” the report said.
“The compiled material included a roster of U.S. personnel scraped from captions on public military photographs; publicly accessible ship and aircraft transponder identifiers; commercial satellite-imagery query scripts; and an inventory of public websites that exposed U.S. naval movements,” the report said.
Most of the report, however, details the misuse of the advanced intelligence technology for China’s military, defense and security, and intelligence purposes.
Claude is a family of advanced large language models and conversational AI assistants created by Anthropic. In addition to its public and subscription services, the leading AI company also conducts classified work for the Pentagon and U.S. intelligence agencies.
Improper Chinese military use of the commercial AI technology involved work on an anti-torpedo fire control system by a Chinese defense company, development of PLA electronic warfare and counter-air defense systems, and information on directed energy weapons for use in a briefing by the PLA for Chinese Communist Party and military leaders, stated the report made public Thursday.
The widespread international misuse of Claude violates Anthropic’s terms of service, and the report suggests the activity violates some U.S. security and fraud laws.
Separately, the FBI, National Security Agency and Cybersecurity and Infrastructure Security Agency reported Tuesday that Chinese AI companies are stealing U.S. AI technology through malicious “distillation” — systematic extraction of restricted proprietary functions and capabilities of the leading U.S. AI models.
The Anthropic report said the weapons development by China and other nations was shut down by banning all accounts linked to the activity and notifying affected industries.
“Across these cases, the actors used Claude to build and refine software for weapons hardware and firmware with which they already had expertise and to which they had access,” the report said.
The operations involving the improper use of Claude were divided among multiple sessions to hide the weapons work and employed covert methods in seeking to avoid detection, the report said.
The Chinese undersea warfare actor used Claude to build parts of an anti-torpedo weapons system and test it online.
Anthropic prohibits the use of Claude for weapons design and development.
Claude was also detected being used by Chinese hackers in China’s Hunan province working for security companies that U.S. officials say are often contracted by Beijing’s intelligence and security services for offensive operations.
The group used Claude for a “coordinated offensive program” of intrusion attempts, reconnaissance of foreign-government networks, vulnerability research, malware development and “an intelligence-collection platform,” the report said.
In Yemen, unspecified threat actors in the northern part of the country used Claude to develop a guided rocket system, a multi-stage ballistic missile and a hypersonic glide vehicle, the report said. The effort included using the AI agent for adapting auto-pilot technology for missiles.
A Chinese state security bureau also used Claude to produce an internal AI-powered surveillance manual and to boost spying on dissidents, including ethnic Uyghurs, Catholics, Tibetan Buddhists, Falun Gong members and Taiwanese Christians.
The operations were aligned with the CCP’s United Front Work Department as part of what the party calls “stability maintenance” surveillance and transnational repression of dissidents, the report said.
Another case involving improper Chinese use of Claude involved Beijing’s “public opinion monitoring” of dissidents, ethnic minorities, diaspora communities and foreign media. The goal of that AI misuse was muting criticism of China through the “three warfares” doctrine — legal warfare, propaganda warfare and psychological warfare — key PLA information weapons.
Russian misuse of Claude centered mainly on intelligence and information operations, targeting Ukraine and Eastern Europe.
One significant operation by a Russia-based threat actor was the use of Claude to produce “first-person view” kamikaze drone swarm software.
“The actors designed the platform for autonomous lethal engagement; the onboard model could select targets (including a “person” target class) and issue detonation commands without a human in the loop,” the report said, identifying the developers as a joint military-civilian team.
The drones were to be used against Ukraine, the report said.
Iran also used Claude for conducting influence operations in support of Tehran’s policies and to counter political opposition groups and Israeli officials and Jewish diaspora groups.
The report concludes that AI is expanding the capabilities of foreign adversaries’ militaries and intelligence and security services in building weapons and population control systems.
“As models become increasingly capable, their risks will increase, unless AI developers and society’s defenders act to make them safer,” the report said.
Other nations linked in the report to improper AI use include Turkey, France, Kenya, United Arab Emirates, Mali and Bangladesh.
Contact the author
Please read our comment policy before commenting.