A cybercriminal utilized AI agents to steal approximately 600,000 credit card records from online stores. The campaign, which began no later than July 2026, remained active as of September 22, 2026. Security firm Gambit reports that the majority of the stolen data originated from just two companies, while malicious skimmers were deployed on the sites of five others.
The attacker employed three specific AI platforms: Strix for vulnerability scanning, Cairn for exploiting systems, and Hermes for post-exploitation and tactical decision-making. Hermes is based on the claude-opus-4.6 model. Between August 23 and August 31, 2026, Strix was executed 146 times across 138 hosts, totaling 633 hours of scanning activity.
The operator, who appears to use Chinese for instructions, successfully compromised at least 119 websites. Targets included a Fortune 500 hospitality firm, a major US airline, an industrial goods distributor, and a fashion retailer. Techniques involved injecting malicious code into JavaScript files, Google tag blocks, and manipulating server-side caches or Kubernetes instances.
The cost of the operation ranged from $12,000 to $18,000, with an average expenditure of approximately $25 per target. Data from the OpenRouter marketplace indicates that the attacker spent $7,000 over a four-week period ending August 25, 2026. The AI agents were also instructed to delete original card data from databases, causing operational failures for several retailers.
