AI-Driven Skimming Campaign Steals 600,000 Credit Card Records

A sophisticated cyberattack campaign using autonomous AI agents has compromised at least 119 websites, stealing more than 600,000 credit card records.

Researchers at the cybersecurity startup Gambit discovered that a financially motivated threat actor has been using open-source AI frameworks to target hundreds of online retailers at scale. The campaign has been active since at least July 2026 and remains ongoing.

Autonomous AI Frameworks Automate Attacks

The attack chain is powered by three distinct AI-driven tools designed to automate different stages of the breach. Strix serves as a penetration testing framework for scanning and vulnerability discovery. Cairn acts as an autonomous exploitation engine, tasked with objectives such as obtaining administrative access or a command shell.

The third component, Hermes, is used for campaign orchestration and tactical decision-making. Hermes utilises the Claude-Opus-4.6 model to direct malicious activity and manage post-exploitation work. Gambit researchers noted that the human operator provides brief instructions to the agents and then allows them to execute the operation autonomously. The operator appears to be based in China.

Broad Targeting and Injection Methods

The campaign has successfully breached large organisations, including a Fortune 500 hospitality company, a major US airline, and a large US industrial supplies distributor. To identify high-value targets, the attacker used website traffic-ranking services to prioritise retailers running custom software, which are often more vulnerable to exploitation.

Once access was obtained, the threat actor deployed credit card skimmers using various techniques. These included appending malicious code to legitimate JavaScript files, adding script tags to checkout pages, poisoning S3 or CDN content, and modifying database fields. In some instances, the attacker even altered Kubernetes deployments or used cron jobs to ensure the skimmer remained active after attempted removals.

Low Costs Enable Scalable Cybercrime

The automation provided by these AI tools makes large-scale attacks highly affordable. Gambit researchers estimated the total cost of the campaign to be between $12,000 and $18,000, representing a marginal cost of roughly $25 per targeted company.

This low cost of entry allows even relatively unskilled actors to launch high-frequency, automated attacks. In many cases, the AI agents were able to gain access to target systems within just a few hours of receiving instructions from the operator.

Data Loss from Automated Cleanup

The investigation also highlighted unintended consequences for victimised retailers. To hide their tracks, the attacker instructed the Hermes agent to perform cleanup procedures after exfiltrating data.

One specific instruction found in the agent’s skill files commanded the AI to “wipe the source fields in batches” after downloading all card data. This automated deletion of database records caused significant operational disruptions and data loss for several retailers during the attack.