AI agents are now hacking computer. They are getting better in all stages of cyber attacks, much faster than most of us. They can combine various aspects of cyber operation together, and can hack autonomously on computer speed and scale. It is going to change everything.
In summer, hackers proved this concept, the industry made it institutional and criminals implemented it. In June, AI company XBow took Top position In a few months, after submitting more than 1,000 new weaknesses on Hackerone’s American leaderboard. In August, seven teams will compete in Darpa’s AI Cyber Challenge Collectively found 54 new weaknesses in the target system, in four hours (of calculations). Even in August, Google Announced Its Big Sleep AI found dozens of new weaknesses in the open-source projects.
it gets worse. In July, CERT of Ukraine discovered a piece of Russian malware that used LLM to automate the cyber attack process, which produced both system reconnaissance and data theft commands in real time. In August, Anthropic reported that he interrupted a dangerous actor who used the AI model of Cloud, Anthropic Automated Complete cyber attack process. It was an impressive use of AI, which entered the network reconnaissance, network and collected the credibility of the victims. AI was able to find out which data to steal, how much money to withdraw from victims, and how to write forcibly recovery email.
