Malicious AI Agents Steal 600,000 Credit Cards And Infect Over 100 Websites
By 813 Staff
Engineers and executives are reacting to Malicious AI Agents Steal 600,000 Credit Cards And Infect Over 100 Websites, according to BleepingComputer (@BleepinComputer) (in the last 24 hours).
Source: https://x.com/BleepinComputer/status/2102795429377192414
Malicious AI agents have been used to compromise more than 100 websites and steal roughly 600,000 credit card numbers, according to reporting from BleepingComputer (@BleepinComputer) published September 23, 2026. The campaign represents one of the first documented cases in which autonomous software agents, rather than human operators working manually, carried out the end-to-end work of compromising sites and deploying payment skimmers at scale. Internal documents reviewed by the outlet describe a toolchain that identifies vulnerable storefronts, injects obfuscated JavaScript into checkout pages, and rotates infrastructure faster than traditional takedown efforts can respond.
Engineers close to the investigation say the agents operated with minimal human direction. Once pointed at a target list, the system appears to have handled reconnaissance, exploit selection, and payload placement on its own, then harvested captured card data and exfiltrated it through disposable relay domains. The 600,000 figure reflects cards observed in command-and-control logs, though investigators caution that the true total could be higher and that some records may be duplicates or test data. The 100-plus affected sites range from small merchants to mid-sized retail platforms, with victims identified across North America and Europe. The rollout of the skimmers has been anything but smooth for defenders: security teams report that signatures written for earlier variants failed to detect the newer, AI-generated obfuscation within hours of deployment.
Why this matters goes beyond the raw number. Skimming attacks have historically required patient human attackers, which naturally limited their scale. Agentic automation removes that bottleneck. If a single operator can direct software to compromise hundreds of sites in days, the economics of payment fraud shift decisively toward the attacker, and small merchants without dedicated security staff absorb most of the risk. The incident also raises unanswered questions about liability, since the agents allegedly chained together known vulnerabilities rather than exploiting a novel zero-day, meaning many of the affected sites could have been patched before the campaign began.
What happens next remains contested. Payment processors and card networks are expected to push accelerated reissuance for exposed cards, while incident responders continue mapping the infrastructure behind the operation. Attribution is unconfirmed; no group has claimed responsibility, and researchers have not publicly linked the campaign to a known threat actor. Regulators in at least two jurisdictions have begun preliminary inquiries, according to people familiar with the matter, though no formal actions have been announced. For now, merchants running outdated checkout software remain the most exposed, and the tools that enabled this campaign are not going away.
Source: https://x.com/BleepinComputer/status/2102795429377192414
