2024 · G - Wholesale and retail trade · Mass Magecart-style digital skimming of online stores
Analysed 2026-11-06 by Hermes
Open this case in the tool →Mass Magecart-style digital skimming of online stores
Expertise to Execution
Evidence coverage: 3/6 stages, weight 16/60
Decisive stage: entry (clearest evidence) · Alternative: Strategy to Execution
(1) Online stores that process card payments on a mainstream e-commerce platform are expected to keep that platform patched and their checkout pages monitored for tampering; for Magento/Adobe Commerce the relevant flaws of summer 2024 (led by CVE-2024-34102, published 11 June 2024) had published patches, vendor guidance and key-rotation procedures well before the August 2024 campaign. (2) In practice the skimmer loaders were planted on a few hundred stores that Malwarebytes judged likely exploited by the same vulnerability, while Sansec measured that 5% of all Adobe Commerce/Magento stores ended that summer skimmer-infected, 75% of the install base unpatched when automated scanning began, the stores themselves blind to the injected code (detection came from Malwarebytes telemetry and Sansec, who contacted all 4,275 affected merchants), and even patched stores left exposed because stolen secret crypt keys were not invalidated, so competing groups kept re-hijacking the same stores. (3) Across the victim population the remedy sat with technical specialists and was published, yet was not realised in day-to-day operations - patching skipped, skimmer monitoring absent, key rotation not adopted: material origin, execution break, so the classification rests on the entry stage (load-bearing, clearest evidence), with the detection and response gaps pointing the same way; no source documents victim leadership deferring, underfunding or discounting these needs, so a leadership-side type is unsupported. The entry vector is not individually confirmed for this specific campaign (Malwarebytes says the actors likely used the same vulnerability), but the victim-side patch/monitoring failure is documented at platform scale and carries the classification. (4) RM18 (Expertise to Execution lens): the exploited Magento flaws were known and remediable - patch available since June 2024 - and execution lagged at platform scale, with 75% of the install base unpatched when the mass scanning began, which enabled the infection of the stores (causal role: Enabling). RM34 (Expertise to Execution lens): the stores' own monitoring never surfaced the injected checkout code, which Malwarebytes describes as effectively invisible without traffic inspection or Developer Tools, so detection had to come from an outside vendor's telemetry - available detection capability was not configured or maintained on the victim side (causal role: Obscuring). RM38 (Expertise to Execution lens): after the intrusion the eradication experts prescribed (rotate secret keys, reset admin passwords, drop unknown API integrations) was not completed by every merchant, so stolen-key persistence survived patching and multiple groups repeatedly re-took the same stores (causal role: Amplifying).
RM18 — Vulnerability / patch latencyRM34 — Monitoring / detection / alert-handling gapRM38 — Containment / eradication failure during responseDiffuse-victim campaign: the row covers hundreds of unnamed online stores on the Magento platform (receiver_country: Canada; Europe (region); Not available), so Country is left null per the method spec for unspecified multi-organisation campaigns; Sector is set to G Wholesale and retail trade because the victim population is predominantly online retail stores, though the population also includes e.g. a university. The two example victims in the source (a European beer manufacturer, a Canadian university) are not named in the Malwarebytes article. Malwarebytes does not name the exploited vulnerability (only that the compromises looked similar and likely used the same vulnerability); the placement in the documented CosmicSting campaign family (CVE-2024-34102, with an August 2024 CNEXT-chaining escalation per Sansec) rests on platform, timing and method and is flagged as an inference, which is why the classification claim is pattern-level. Sansec's figures cover all seven CosmicSting groups in summer 2024 (4,275 stores), a superset of the few hundred stores in this EuRepoC row. No individual store's disclosure is documented, but absence of notification news was not coded as RM39 - no positive evidence of a notification failure. Unattributed actor does not lower confidence (victim-side assessment). Unattended run: no action required user consent.