Hackers behind the Chinese government were found to have used the artificial intelligence (AI) servi..

Hackers behind the Chinese government were found to have used the artificial intelligence (AI) service "Claude" of the US company Anslogic to break into foreign companies and government agencies in September. Following Russia's attack on Ukraine using Google AI, the 'cyber cold war' using AI as a tool is intensifying.

According to the Wall Street Journal (WSJ) on the 13th (local time), Chinese hackers have been using AI for individual tasks such as writing phishing e-mails and exploring vulnerable systems for years, but in recent attacks targeting dozens of targets, 80-90% of automation has been made to the extent that only a small part of human intervention is involved.

The hackers carried out the attack with a click of a button, requiring only minimal human intervention such as 'yes, go ahead' and 'don't go ahead'. Anthropic blocked the attack and blocked the hacker's account, but before that, it succeeded in breaking into up to four cases. In one case, hackers also succeeded in querying internal databases and extracting data by instructing Anthropic's Claude AI tool.

Last summer, cybersecurity firm Bollexity spotted Chinese-backed hackers using AI tools to automate some of their hacking operations against companies, research institutes and non-governmental organizations. Bollexity President Stephen Adair explained that hackers used a large language model (LLM) to decide how to select targets, write phishing emails, and write malicious codes. "AI is giving hackers the ability to do more things faster," Adair said.

Last week, Google also announced that hackers linked to the Russian government attacked Ukraine using Gemini AI, which generates customized malicious code instructions in real time.

A spokesman for the Chinese Embassy in Washington, D.C., accused the U.S. of using cybersecurity to slander and slander China. "China firmly opposes and cracks down on all forms of cyberattacks," he said.

Anthropic did not disclose a list of companies and government agencies that hackers attempted to infiltrate, but said it had detected 30 targets. Some hacks have succeeded in stealing top-secret information.

In the meantime, hackers often use open-source code for hacking. This is because open-source codes can be written for free and can be modified to disable malicious activity restrictions on AI services. But in order to use Claude in the attack, Chinese hackers diverted the safeguards of Anthropic in a way called "jailbreak." The hackers even tricked Claude into conducting security checks on behalf of the target agency.

However, Anthropic explained that automated hacking does not have complete autonomy, causing errors due to "perception." Referring to some hacking attempts, Klein explained, "Even though we haven't actually accessed it, we can exaggerate our capabilities and report that we have been able to access our internal systems."