Chinese Conduct Cyber Hack With AI

Chinese-sponsored actors are responsible for an artificial intelligence cyberattack against about 30 entities. Experts are calling the hack the first documented case in which a cyberattack was executed with very little human intervention.  

The hack “represents multiple firsts in AI-enabled threat actor capabilities,” according to a report by Anthropic, an AI safety and research company, detailing the findings of the attack.  

In the middle of September, AI experts say they detected “a highly sophisticated cyber espionage operation conducted by a Chinese state-sponsored group we’ve designated GTG-1002 that represents a fundamental shift in how advanced threat actors use AI.” 

The investigation uncovered a “well-resourced, professionally coordinated operation involving multiple simultaneous targeted intrusions,” Anthropic reports. The 30 or so targets included financial
institutions, chemical manufacturing companies, major technology corporations, and government agencies in various countries. Through AI, the hackers succeeded in making a “handful of successful intrusions.” 

China, Iran, North Korea, and Russia represent the greatest cybersecurity threats to the U.S., according to America’s Cyber Defense Agency, with China representing “persistent cyber threats.”  

Chinese-sponsored hackers used the AI technology known as Claude, which is similar to ChatGPT and was developed by Anthropic, to conduct the attack this fall.  

Despite Anthropic developing safety and security measures to prevent the abuse and misuse of its AI technologies, hackers successfully found a way around the safeguards.  

Hackers manipulated “Claude Code to support reconnaissance, vulnerability discovery, exploitation, lateral movement, credential harvesting, data analysis, and exfiltration operations largely autonomously.”  

It was discovered that the threat actor was “able to leverage AI to execute 80-90% of tactical operations” independent of human intervention.  

Acting on its own, the AI found “vulnerabilities in targets selected by human operators and successfully exploited them in live operations, then performed a wide range of post-exploitation activities from analysis, lateral movement, privilege escalation, data access, to data exfiltration.”  

While Anthropic has expected AI to continue to evolve, “what has stood out to us is how quickly they have done so at scale,” according to the report. Still, the hack was not flawless, with Claude overstating, or even fabricating, findings to the hackers. These and other flaws in AI technology currently pose an obstacle to “fully autonomous cyberattacks,” according to the report.  

By sharing their findings following the investigation into the hack, Anthropic says it hopes to strengthen defenses against AI abuse.  

The cyberattack has “substantial implications for cybersecurity and underscores the urgent need for AI safeguards,” Anthropic says. “We plan to continue releasing reports like this regularly, and to be transparent about the threats we find.”