AI Agents Hit 100 Online Stores for Cheap and Walked Off With 600,000 Stolen Credit Cards

A Chinese-speaking attacker turned a handful of off-the-shelf AI agents loose on online retailers and walked away with more than 600,000 credit card records. The whole operation appears to have cost about $8,000.

Quartz reported that the attacker rented autonomous AI agent frameworks and pointed them at up to roughly 100 online companies, paying around $25 per target. Researchers at Gambit Security reconstructed the campaign after the operator accidentally left staging infrastructure sitting on the open web.

The agents handled work a human crew used to do by hand. One tool, Strix, hunted for vulnerabilities. Another, Cairn, ran the exploitation. A third, Hermes, orchestrated the campaign across dozens of targets at once.

What sat behind those agents is the detail worth pausing on. The operator leaned on the Chinese models DeepSeek and Kimi, plus an older Claude Opus 4.6. Newer Claude releases refused the criminal prompts, and Anthropic banned the account tied to the activity.

The campaign has been running since at least July. CyberInsider reported a concentrated burst between September 10 and 15, with about 105 attack projects and at least 27 organizations compromised to some degree.

The totals carry a caveat. The "up to 100" companies and "600,000 plus" cards are Gambit's reconstruction from that exposed staging server, not a universe every victim has confirmed. Reported targets span retailers, travel companies, and industrial firms, and coverage points to at least one Fortune 500 name.

The pattern mirrors what defenders saw when AI agents helped hit 395 organizations running PaperCut, and it lands on the same soft target as the StyleSmuggler skimming campaign against Adobe Commerce stores: the checkout page, where card data lives.