Researchers at Gambit Research disclosed on September 24, 2026, a sprawling e-commerce campaign in which autonomous AI agents systematically stole more than 600,000 valid credit card numbers and planted payment skimmers on at least 119 websites. The operation has been active since at least July and was still running on September 22, according to their telemetry. A human operator — likely Chinese-speaking — sets brief goals; agents then handle scanning, exploitation, and persistence on their own.
The campaign revolves around three named tools. Strix performs reconnaissance: between August 23 and 31 it ran 146 jobs against 138 hosts, consuming 633 scan-hours. Cairn drives autonomous exploitation when vulnerabilities appear. Hermes orchestrates the workflow through the claude-opus-4.6 model under a persona called "SOUL - Red Team Operator" with 121 defined skills. The operator spent roughly $7,000 via OpenRouter over four weeks; Gambit estimates total AI spend at $12,000–$18,000 — averaging $25.46 per attack target.
In five days alone, agents executed at least 100 attacks against more than 27 companies. Two victims yielded 600,000-plus verified card numbers; five others received skimmers embedded in checkout flows. Victims span Fortune 500 hospitality brands, a major U.S. airline, an industrial distributor, and a fashion retailer — showing the actors prioritized high-volume transaction sites, not small niche shops.
Injection methods vary and reveal deep web-stack knowledge: JavaScript appends in theme files, extra `<script>` tags in checkout, manipulation of Google Tag Manager blocks, poisoning of S3 and CDN objects, writes directly into database fields, compromise of Kubernetes clusters, and cron jobs that restore skimmer code after cleanup. In at least one Magento case, attackers wiped card fields after exfiltration — causing operational disruption beyond the data breach and forcing victims into urgent backup recovery.
For consumers, the skimmer behaves like classic card theft: payment data is captured at checkout in real time. The difference is scale. One operator with AI agents can map hundreds of e-commerce stacks in parallel, choose injection points, and iterate payloads without hand-writing every exploit. That lowers the bar for massive cybercrime and challenges traditional WAF signatures built around human attack tempo.
Gambit's analysis matters for Nordic retailers and payment providers: Magecart-style skimmers are not new, but AI orchestration compresses detection windows. Security teams should monitor checkout file integrity, CSP violations, and anomalous CDN changes with the same urgency as server RCE patches.
Evidence Rail
- Confirmed: Gambit Research report; 600,000+ cards from two companies; 119 websites with skimmers; July–September 22 activity; Strix/Cairn/Hermes tooling documented.
- Reported: Chinese-speaking operator; OpenRouter spend ~$7,000/4 weeks; mean cost $25.46/target; Magento fields wiped post-exfiltration.
- Unconfirmed: Operator's full identity; whether stolen cards already sold in bulk on dark web markets.
What affected consumers and merchants should do now
- Consumers: review card transactions from July 2026 onward; request replacement cards if suspicious; enable transaction alerts.
- Merchants: audit checkout JS, themes, and CDN objects for unknown scripts; verify integrity hashes daily.
- Payment providers: warn merchant customers in affected verticals; offer forced re-auth for stored cards.
- Everyone: report suspected skimmers to police and PCI contacts; preserve logs for potential DFIR.
In Brief
- AI agents stole 600,000+ cards and infected 119+ retail sites in weeks, not months.
- The Strix/Cairn/Hermes chain cost the attacker roughly $25 per target on average.
- Skimmers returned via cron and CDN poisoning — checkout file integrity is critical.