Besides its negative impacts on the environment and resources, artificial intelligence (AI) is rapidly becoming a high-risk tool in the field of cybersecurity. As evidence, OALABS Research recently exposed a case where an amateur Ethiopian hacker used AI agents from Claude and the OpenAI Codex to carry out a series of cyberattacks.
Despite the command being full of spelling and grammatical errors and offering only vague instructions, this individual successfully compromised numerous private servers. This hacker infiltrated the secure data of at least 14 companies and even plotted to steal $4 million worth of cryptocurrency.

Amateur hackers used Claude and the Codex to attack 14 businesses.
Hackers and their tactics of exploiting AI to bypass system barriers.
The attackers bypassed the robust security measures of the Claude Opus model relatively easily through psychological manipulation. By posing as members researching security vulnerabilities, the hackers tricked the AI chatbot into automatically providing the entire source code of the attack. The system even outlined detailed methods of exploitation such as extortion, data sales, and directly instructed how to steal data. This chatbot only blocked one request that deliberately targeted the personal digital accounts of a specific family.
This unusual incident was fully exposed after a victim whose server was compromised contacted OALABS experts. Notably, the identity of the young hacker was revealed simply because he had previously requested the AI to edit his personal profile. The incident raised deep concerns about the public's easy access to these powerful technological tools without requiring specialized knowledge. Users could easily use another AI intermediary to draft similar dangerous attack commands.
Currently, tech giants like Anthropic and OpenAI are at a standstill in finding a definitive solution to prevent such attacks. Tightening the model's limitations would directly deprive genuine security researchers of the opportunity to strengthen the system. Conversely, maintaining the status quo would leave the line between ethical use and illicit gain as an insurmountable obstacle for AI. Finding the optimal solution to prevent malicious user intent remains an unsolved challenge.
According to Thanh Nien Newspaper
Source: https://baoangiang.com.vn/hacker-nghiep-du-dung-chatbot-ai-tan-cong-14-cong-ty-a490360.html










