On October 27, CISA added CVE-2026-48356 to its Known Exploited Vulnerabilities catalog and gave federal agencies until November 10 to fix it. Adobe had shipped the patch back in July. The gap between those two dates is most of the story here, and it is not a comfortable one for anyone running a Magento store.
The flaw carries a nickname, PolyShell, and the nickname tells you how it works. Adobe classifies it as an unrestricted upload of a file with a dangerous type (CWE-434), the kind of weakness where an application accepts a file it had every reason to reject. The NVD scores it 9.3 out of 10. Adobe's own bulletin rates it 9.6. The difference between those two numbers comes down to how much each group thinks an attacker can disrupt availability, which is a detail worth about as much attention as it sounds like. The first digit is what matters. This is a critical, remotely reachable bug in some of the most widely deployed online store software on the internet.
What a polyglot file actually is
A polyglot file is one that is valid as more than one format at once. The same bytes can be read as a perfectly ordinary image by a picture viewer and as executable PHP by a web server. Upload filters that only inspect the first few bytes of a file, the part that announces "I am an image," see a legitimate image header and let it through. Something downstream then hands the same file to the PHP interpreter, which reads past the header and runs the code hiding behind it. That is the move the PolyShell name points at, and it is an old one. Polyglot tricks have been used against image boards, forums, and avatar uploaders for years. Dropping the same technique into the file-handling path of a commercial storefront is what turned a familiar nuisance into a payment-data problem. What changed is where it landed.
Adobe's advisory says exploitation requires some user interaction, and the security firm Sansec, whose research several write-ups lean on, traces the upload path to functionality reachable before a shopper ever logs in. Those two statements are not in conflict. User interaction in the scoring sense can mean something as mundane as a staff member later viewing an affected page. The practical effect is that an attacker who has never authenticated to your store can get a file onto your server and, with the right follow-up, get that file executed.
Once code runs on the server, the attacker is no longer poking at a web form. They are inside the application, with the same reach the store itself has: the database of orders and customers, the admin accounts, the payment integration. A web shell planted this way is a durable foothold. It does not depend on the original bug staying open.
Patched in July, exploited since March
Here is the part that makes this more than a routine patch notice. Adobe fixed PolyShell in security bulletin APSB26-73 on July 14, with isolated patches documented separately in August. But Sansec reports exploitation attempts stretching back to March, including a mass wave on March 30 in which attackers dropped web shells and injected JavaScript into store pages. That is roughly three and a half months of a working attack against a bug that had no public patch yet.
Adobe's bulletin, as of its last revision, still carried the line that the company was "not aware of any exploits in the wild." That wording predates the KEV listing, and CISA does not add things to that catalog on a hunch. When the two disagree, the KEV entry is the one that reflects what defenders are actually seeing. One report counts on the order of 950 compromised stores tied to this flaw, a figure that comes from a single source and should be treated as an estimate rather than a census, but even discounted heavily it describes a campaign, not a proof of concept.
The most recent activity, from late October, loads attacker-controlled scripts out of a store's own content blocks. That detail matters more than it first appears, and it leads directly to the thing most people are about to get wrong.
Why patching is the easy half
Applying APSB26-73, or the matching isolated patch, closes the door. It does not evict anyone already inside. If an attacker uploaded a web shell in April and injected a loader into your CMS blocks in October, upgrading your Magento build in November stops new uploads and does nothing to the shell sitting in your media directory or the script quietly skimming your checkout page. I have watched teams declare victory on the version number and never look at the file system. That rule sounds pedantic until the first incident that started weeks before anyone patched.
So the cleanup has a shape. Apply the fix first, whether that is the 2026-jul release for your product line or the isolated patch, and confirm it with Adobe's version tooling rather than trusting the ticket that says it was done. Then go looking for what might already be there. Sansec's guidance points at specific places: unexpected files in pub/media/custom_options/, including images that are secretly PHP, and a backdoor named accesson.php turning up in new subfolders under assets/images/. Search your cms_block and cms_page tables for injected scripts, since that is where the October campaign is hiding its loaders. Deny web access to the custom options directory at the web server level, which limits what a planted file can do even if you miss it. Review admin accounts for ones you do not recognize. And assume that database, admin, and API credentials reachable from a compromised store are burned, so rotate them after you have cleaned up, not before.
The affected versions are broad. Adobe lists Commerce 2.4.9 and the recent patch levels of 2.4.8 down through 2.4.4, the B2B builds, Magento Open Source on the same lines, and Commerce Events from 1.6.0 through 1.20.0, fixed in 1.21.0. There is a minor disagreement in the sources about whether the May 2026 cut of 2.4.9 already contained the fix, with Adobe listing 2.4.9 as affected and Sansec suggesting otherwise. When a vendor's affected list and a researcher's note diverge, patch to the vendor's list. The cost of an unnecessary upgrade is an afternoon. The cost of trusting the optimistic reading is the incident.
The deadline is a signal, not a mandate
CISA's November 10 due date binds federal civilian agencies. Almost nobody running a Magento store is a federal civilian agency. It would be easy to read the KEV listing as somebody else's problem. That reading misses the point of the catalog, which is not a compliance instrument for most of the people who should act on it. It is the clearest public signal available that a given flaw is being used right now, against real targets, by people who already have working tooling. The deadline exists for the agencies. The warning is for everyone.
What I would check first, if I ran or advised a store on any affected version, is not the patch status. It is the file system and the CMS tables, because the exploitation window opened in spring and the patch did not start the clock. The version number tells you whether you are still vulnerable to new attacks. The media directory tells you whether the old ones already succeeded. Those are different questions, and only one of them is answered by an upgrade.
References
Rescana. (2026, October). Adobe Commerce and Magento CVE-2026-48356 (PolyShell) added to CISA KEV: Apply APSB26-73 and check for planted files. Rescana. https://www.rescana.com/post/adobe-commerce-magento-cve-2026-48356-file-upload-rce-kev-october-2026
Aviatrix Threat Research Center. (2026, October 28). Adobe Commerce PolyShell & Node2Shell Brotli RCE: 2026 breach analysis. Aviatrix. https://aviatrix.ai/threat-research-center/adobe-commerce-polyshell-attacks-breach-hundreds-of-stores-as-brotli-exploit-traffic-emerges/