A large payment-card theft campaign is showing how AI agents can move beyond reconnaissance and coding support to perform substantial portions of a real intrusion. According to research reported by BleepingComputer, an attacker combined three agent-based tools to discover vulnerable retailers, obtain access and deploy web skimmers across compromised infrastructure.
Automation Across the Attack Chain
The operation reportedly used Strix for scanning and vulnerability discovery, Cairn for autonomous exploitation, and Hermes for orchestration and post-compromise decisions. Instead of manually directing every command, the human operator supplied relatively short objectives and allowed the tools to perform extended runs against selected targets.
Researchers observed more than 100 attack waves over a five-day period, with varying levels of success against at least 27 companies. Evidence collected during the investigation connected the campaign to skimmers placed on at least 119 websites and more than 600,000 valid payment-card records taken from two businesses.
The attacker adapted the installation method to each environment. Techniques included modifying legitimate JavaScript, inserting scripts into checkout pages, poisoning content delivered through storage or content-distribution services, changing database fields and altering Kubernetes deployments. Persistence mechanisms were also used to restore skimming code after removal.
Why This Changes the Economics of Cybercrime
The campaign's estimated operating cost was only a few dollars to several dozen dollars for each target. That matters because agentic automation allows criminals to test far more organizations without building a large intrusion team.
I believe the primary warning is not that AI has created unfamiliar vulnerabilities. The agents succeeded by finding and combining existing weaknesses in custom applications, cloud permissions, deployment systems and website administration. Their advantage was speed, consistency and the ability to keep trying.
Defensive Priorities
- Monitor checkout scripts and critical JavaScript files for unauthorized changes.
- Restrict write access to content-delivery, storage and tag-management systems.
- Use file-integrity monitoring across web servers and application containers.
- Review Kubernetes changes, scheduled tasks and database modifications.
- Prepare for destructive cleanup that may erase source data after theft.
Retailers should assume that attackers can now automate target selection and exploitation continuously. Defenses must become equally continuous, with rapid change detection and automated validation focused specifically on payment infrastructure.