AI-Powered Campaign Targets Hundreds of Online Retailers

* The campaign utilized three AI harnesses for automation: vulnerability research, exploitation, and orchestration.
* 105 attack projects were launched between September 10 and 15.
* At least 27 companies were compromised to varying degrees.
* Information from over 600,000 unexpired credit cards was stolen from two victim companies, including over 488,000 cards from the US.
* Skimmer scripts were injected into five online stores.
* Access was gained to a Fortune 500 hospitality company and three U.S. firms (airline, industrial supplies distributor, fashion retailer).
* The campaign utilized the open source AI penetration testing tool Strix for vulnerability hunting.
* The agents used included Cairn for launching attack projects and Hermes for orchestration and hacking activities.
* 1,951 prompts were typed by a human across 260 sessions.
* The mean cost estimated for completed scans was $25.46 over 101 scans.

The narrative demonstrates a shift from traditional targeted attacks to highly scaled, agent-driven operations that leverage readily available open-source AI tools to achieve significant financial outcomes at marginal operational cost. The core implication lies in the automation of reconnaissance and exploitation, transforming what was once a labor-intensive process into an autonomous system capable of dynamic TTP selection across multiple targets simultaneously. The focus shifts from vulnerability discovery to persistent, self-directed execution orchestrated by memory-retaining agents. This challenges traditional security models predicated on human decision-making and attribution, as the attack vector is defined by agent protocols rather than discrete human commands. The low cost associated with utilizing open source tooling suggests that access to sophisticated offensive capabilities is becoming democratized, lowering the barrier to entry for financially motivated groups while simultaneously increasing the velocity and scope of potential harm against an uncoordinated defensive posture.
* BRIDGE QUESTIONS: If autonomous AI agents can execute multi-stage attacks at a marginal cost, how must defense strategies fundamentally reorient themselves away from signature-based or per-incident response toward monitoring and disrupting the underlying agentic decision-making loops? What are the geopolitical and economic consequences when attack infrastructure costs plummet due to open-source AI adoption? Does the reliance on these agents redefine the concept of human agency in cybersecurity, where operational capability is divorced from direct human control?