From First Probe to 471 Stores: The PolyShell Timeline and Its Lessons
From First Probe to 471 Stores: The PolyShell Timeline and Its Lessons The public record for CVE-2026-48356 shows how quickly an upload defect in a common commerce platform can become an operational incident. Sansec pu
From First Probe to 471 Stores: The PolyShell Timeline and Its Lessons
The public record for CVE-2026-48356 shows how quickly an upload defect in a common commerce platform can become an operational incident. Sansec published its PolyShell research in March 2026, observed a mass attack wave later that month, and continued to document storefront compromises in October. Adobe published a critical advisory and fixed builds in July.
The technical starting point was Magento's guest-cart custom-option flow. A request could include base64 file data, a filename, and a MIME type. Sansec found that the server did not verify the option ID against the product, did not require a file-type option, and did not restrict dangerous extensions. The file was written under pub/media/custom_options/quote/. The image-header check could be bypassed with a polyglot.
The first tactical lesson is that public exposure did not require a logged-in administrator. The guest-cart routes used in the attack were available without authentication. Adobe's CVSS 3.1 vector for the CVE is AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:L, with a base score of 9.6. The user-interaction component matters, but it did not make the upload path safe to leave open.
The second lesson is speed. On March 30, Sansec reported 471 stores compromised in one hour through a PolyShell attack wave. The attackers injected an obfuscated JavaScript loader, used localStorage for persistence, and selected responses based on referrer, page title, and query parameters. The campaign was automated enough to create new victims while the researchers were documenting it.
The third lesson is that detection has to follow the post-upload chain. A webshell could drop accesson.php backdoors across directories. The CMS loader could remain in pages and static blocks after the uploaded file was removed. A scanner that looked only for one filename or one directory could miss the database content, and a browser-side scan could miss the server-side persistence.
Adobe's July bulletin lists affected releases and fixed builds. The affected Commerce ranges include 2.4.9 and the 2.4.8-p5, 2.4.7-p10, 2.4.6-p15, 2.4.5-p17, and 2.4.4-p18 families and earlier. Open Source is affected through the 2.4.9, 2.4.8-p5, 2.4.7-p10, and 2.4.6-p15 lines. Fixed builds use the 2026-jul suffix. B2B and Events have separate product rows.
The fourth lesson is that a patch can arrive after exploitation has already started. Adobe's bulletin says the company was not aware of exploits in the wild at publication. Sansec's March research and mass-wave reporting document earlier observed attacks. The apparent difference is a reminder to treat vendor statements as scoped to what the vendor knows at that time, and to combine them with incident evidence from the affected environment.
The October reporting from the source article describes a longer-running campaign with a larger store count. Sansec tied 471 storefronts to PolyShell as an initial access vector and described additional compromises through other upload flaws. The details should not be flattened into a claim that every infected store used CVE-2026-48356. The value of the timeline is the sequence: a file-upload weakness, server-side execution, content injection, and a persistent browser payload.
For defenders, the response has to cover all four stages. Patch the affected application. Block the vulnerable request while the patch is pending. Deny web access and execution under the custom-options upload directory. Inspect the filesystem, CMS content, administrator accounts, and logs for evidence that the path was used.
The exposure context needs the same discipline. ZoomEye's app="Magento" query returned 134,284 product fingerprints, and the exact vul.cve="CVE-2026-48356" query returned zero. Those numbers describe a broad product population and a narrow CVE index result. They do not turn a store into a confirmed victim or rule out an affected deployment.
The final lesson concerns repeat compromise. Sansec described a store on 2.4.7-p10 with all available core fixes that still had an unauthorized administrator account used to insert malicious content. The patch closed the upload route, but the attacker retained another path. Patch completion, account review, content cleanup, and file-integrity checks are separate tasks for a reason.
PolyShell is not a story about one missing file extension. It is a chain of trust failures that turned a guest cart into a file drop, then into a server and browser problem. The practical conclusion is to make the boundaries explicit, verify the deployed state, and keep looking for evidence after the patch is applied.
References
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.