Inside the AI Attack

• Autonomous Chinese AI agents have compromised dozens of retail sites and stolen more than 600,000 payment records since July 2026, with the campaign still active.
• Three low-cost harnesses—Strix, Cairn, and Hermes—enabled attackers to hit about 10 companies a day for roughly $25 each.
• Victims include a Fortune 500 hospitality firm, a major U.S. airline, and other large American retailers running custom software.

By Samuel López | USA Herald

WASHINGTON - Gambit researchers uncovered a live Chinese cyber campaign that has used autonomous AI agents to plant credit-card skimmers on retail websites worldwide and vacuum up payment data.

The operation began in July 2026. Since then the bots have launched hundreds of attack projects, broken into dozens of organizations, and walked off with at least 600,000 payment records. The campaign is still running. Gambit recovered the operator’s staging server, watched the skimmers operating on live victim sites, and pieced the whole effort together from logs and the AI’s own notes.

Between September 10 and 15 alone the agents fired off 105 attack waves and compromised 27 organizations to varying degrees. Named victims include a Fortune 500 hospitality company, a major U.S. airline, a large private industrial-supplies distributor, and a U.S. online fashion retailer. One of the AI tools pulled ranked lists of potential targets and focused on sites running custom-built software.

The attackers appear to be financially motivated Chinese operators. They rely on three AI harnesses—Strix, Cairn, and Hermes—that can handle almost the entire attack chain without constant human guidance. The systems strike roughly ten companies a day at a handful of dollars per target. In four weeks the operators spent about $7,000; total cost for the campaign so far sits near $18,000. That works out to a mean of $25.46 per completed scan, ranging from $3.13 for the cheapest target to $79.31 for the most expensive.

Where the agents gained access, they usually did so in less than a day and often in just a few hours. Playbook instructions recovered by Gambit also authorized data deletion and cleanup routines that can disrupt a company’s operations. Those instructions have already been carried out in some of the breaches. Hermes stands out as an open-source autonomous agent equipped with persistent memory, self-written and self-edited skills, a searchable archive of past sessions, scheduled jobs, and a web console.

On the staging server it ran under a Chinese system persona labeled ‘SOUL – Red Team Operator.’ That persona loaded 121 skills, 78 of them attack skills. The low cost and high speed of the operation show how readily available AI frameworks now let small teams scale digital theft across borders with minimal overhead. Retailers that still rely on custom-built storefronts remain especially exposed while the campaign continues.

Stay ahead of the next breach. For exclusive investigative reporting on cyber campaigns, national-security threats, and the legal battles that follow—subscribe free to the USA Herald newsletter at https://usaherald.com/newsletter.