AI-driven attack wave hits online retailers, swiping over 600,000 card records
A financially motivated attacker is using open-source AI agent frameworks to automate large-scale attacks against online retailers, compromising at least 27 companies and deploying card-skimmer malware on multiple websites. The campaign, active since July, has stolen more than 600,000 valid credit card details, with the AI agents handling scanning, exploitation, and post-compromise actions under brief human instructions. Researchers observed over 100 attack waves in a five-day period, with skimmers injected via various methods including modified JavaScript, poisoned CDN content, and altered database fields.
The attack chain relies on three open-source AI frameworks working in sequence: Strix performs reconnaissance and vulnerability scanning, Cairn executes exploitation to gain access, and Hermes orchestrates the overall campaign using a "SOUL - Red Team Operator" persona with 121 skills. Between August 23 and 31 alone, Strix ran 146 times against 138 hosts, accumulating 633 hours of scanning activity. The human operator, believed to be Chinese-speaking, provided only brief strategic instructions before letting the agents operate autonomously.
The financial scale of the operation is notable: researchers found an OpenRouter account showing roughly $7,000 spent over four weeks, with total estimated costs between $12,000 and $18,000. This translates to an average cost of about $25 per targeted company, making the operation remarkably cost-efficient. The attacker also prioritized targets using a traffic-ranking service, focusing on sites running custom software likely to contain exploitable vulnerabilities.
This campaign could signal a shift in cybercrime economics, where AI agents dramatically lower the barrier to large-scale attacks. Small and mid-sized retailers may face heightened risk as attackers can now target hundreds of companies simultaneously with minimal human oversight. Consumers whose card data was stolen could experience fraud, identity theft, and financial losses. The operational disruptions from wiped databases may also harm retailers' ability to process legitimate transactions. If this model proves successful, other financially motivated actors could adopt similar AI-driven approaches, potentially normalizing automated crime waves across multiple industries.