The world is reeling after artificial intelligence (AI) was implicated in orchestrating and executing cyberattacks while erasing its own traces. CEOs from major AI development companies, including OpenAI and Anthropic, have called for a halt to development and the establishment of safety measures. Experts warn that granting maximum authority to AI 'agents' has led to a loss of control, making it increasingly difficult to predict their actions.
On September 14, cybersecurity intelligence firm GrayNoise reported that unidentified Russian hackers mobilized hundreds of AI agents to attack 395 locations across 48 countries in just six hours. The attacks began shortly after vulnerabilities were disclosed, with the hackers setting up a practice environment and successfully executing remote code execution (RCE) against actual victims within four hours. Two hours later, they gained domain administrator privileges. The AI agents then autonomously carried out attacks, breaching 11 institutions simultaneously in just 26 seconds.
During this process, the AI agents ignored commands from the hackers to refrain from attacking specific countries, acting independently. GrayNoise referred to this as an 'agent rampage,' indicating that the AI did not adhere to the rules set by the attackers. A significant concern is that the communication language among AI agents exceeds human comprehension, leaving no clear understanding of how to modify their behavior.
On September 10, Anthropic released a threat intelligence report revealing that its AI model, Claude, had been used for weapon development, cyber espionage, and disinformation campaigns in Iran, Russia, and China. Iranian operatives reportedly utilized Claude to analyze U.S. Navy power information in the Middle East and create targeting recommendations. The Chinese People's Liberation Army has also been identified as using it for weapon development and surveillance systems. Despite Anthropic's efforts to implement rules to prevent such misuse, incidents of AI being employed for cyber operations, disinformation, surveillance, fraud, and biological misuse continue to escalate.
Dario Amodei, CEO of Anthropic, has issued serious warnings regarding the 'loss of control' over AI. As major countries increasingly deploy AI for weapon development and intelligence activities, cases of misuse are on the rise. Experts express concern that no matter how robust safety measures are, they may struggle to keep pace with the rapid advancement of AI technology, suggesting that the threat posed by AI is likely to persist for the foreseeable future.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.