A major cyber heist has unfolded, with a financially driven threat actor exploiting open-source AI agent frameworks to compromise over 600,000 credit card records from online retailers. The campaign, active since at least July and still ongoing, has targeted hundreds of companies, successfully launching more than 100 attacks and compromising 27 firms in just five days. Notably, the attacker stole card details from two companies and deployed skimmer malware on the websites of five others to harvest payment data. The operation utilizes three AI tools, including Strix and Hermes, which were used extensively to execute the attack chain. Strix alone scanned 138 hosts over 633 hours in late August, while Hermes possessed numerous attack-related skills. Researchers from cybersecurity firm Gambit report that the human operator, believed to be Chinese, provided minimal instructions to the AI agents, allowing them to carry out the attacks autonomously.
The skimmer malware was injected into websites through various methods, such as appending malicious code to JavaScript files, modifying database fields, and using cron jobs to restore malware after removal. In total, at least 119 websites have been compromised, including those of large organizations like a Fortune 500 hospitality company and a major U.S. airline. The attacker prioritized targets using a website traffic-ranking service, focusing on those running custom software, which were deemed more vulnerable. Gambit researchers found the operation cost between $12,000 and $18,000, with an average cost of $25 per target, highlighting the low expense and high efficiency of such attacks. They also discovered that the AI agent was instructed to erase card data from Magento databases post-exfiltration, causing disruptions for several retailers. The investigation underscores the ease with which even less skilled threat actors can deploy these attacks using AI tools, emphasizing the need for robust security measures and plans to mitigate potential data loss from cleanup routines.

