Four bypasses in a fortnight: the PaperCut cycle shows why patched is not a state
watchTowr's 9 October account of the PaperCut NG and MF emergency patch cycle documents bypasses of three successive fixes and a new post-authentication flaw that chains with one of them. The public vulnerability record still disagrees with the vendor on which builds are fixed.

A remediation record that moved four times
On 9 October 2026 watchTowr published a consolidated account of the PaperCut NG and MF vulnerability cycle that began in late August [1]. Across roughly a fortnight, four separate bypasses of vendor remediation were identified, two of them for the same underlying issue, and a further post-authentication remote code execution flaw was found in a build that had been shipped to close one of those bypasses [1].
The starting point was 27 August 2026, when PaperCut published an urgent bulletin confirming active exploitation of its print management software and released emergency patches for customers with public-facing servers [3]. The vendor said it had reproduced the vulnerability using information supplied by a university customer’s security and incident response teams, and that it was aware of confirmed customer incidents [3][5]. No CVE identifiers or severity scores existed yet, and the vendor treated every supported version as potentially affected [12].
Within a day the first fix was shown to be incomplete. Both watchTowr and Huntress independently reported bypasses of the original patches, along with an additional authentication bypass, and passed them to the vendor [4]. Emergency Patch Release 2 followed on 28 August, covering versions 24, 25 and 26 across Windows, Linux and macOS, with the vendor urging customers who had applied the first patch to apply the second as well [4]. Rapid7 put the position plainly: organisations that had applied the first or second emergency patch were not fully protected [5]. A third emergency patch followed on 1 September [5].
watchTowr’s account continues past that point. A bypass of the second authentication fix was identified in the build released on 28 August, a further fix for it shipped on 1 September, and a separately reported post-authentication remote code execution issue was closed in version 26.0.5 on 10 September [1]. The researchers state that the second authentication bypass and the new post-authentication flaw together formed an unauthenticated remote code execution chain against the intermediate build [1].
The flaws, at class level
CVE-2026-81578 is an improper access control flaw in the web management interface. Unauthenticated remote requests to administrative functions can trigger backend actions before access validation completes, which allows certain system configurations to be modified [7]. CVE-2026-82078 is an unsafe reflection issue: database driver classes are instantiated from configurable driver names without allowlist validation, so an actor able to change configuration parameters can execute arbitrary Java bytecode with the privileges of the server process [8]. The first provides the configuration write, the second converts it to code execution, and together they produce unauthenticated code execution [2].
CVE-2026-82077, the flaw closed in 26.0.5, is recorded as a path traversal in a scan-to-fax component that allows an authenticated administrator to execute commands on the underlying host through crafted provider settings [9]. On its own it requires high privileges. Chained behind an authentication bypass it does not.
The reason the first authentication fix failed is instructive without being a method. watchTowr attributes the bypasses to authorisation checks applied inconsistently across pages, so that the same privileged function remained reachable through a route the fix had not covered [1]. That is a class of remediation failure, not a product defect: a fix written against the request that was observed rather than against the authorisation boundary that was broken.
The record disagrees with itself
A defender trying to establish whether a given server is fixed has to reconcile two inconsistencies in the public record. The first is severity. The National Vulnerability Database scores CVE-2026-81578 at 9.8 under CVSS 3.1 as the primary assessment, while the vendor-supplied CVSS 4.0 score for the same flaw is 8.8 [7]. For CVE-2026-82078 the order reverses: 9.1 under CVSS 3.1 and 9.4 under CVSS 4.0 [8]. Neither set is wrong. They are different scoring systems applied by different parties, and an organisation whose prioritisation threshold sits between 8.8 and 9.8 will reach a different triage decision depending on which feed it reads.
The second inconsistency is more consequential. Vendor-supplied version data for both flaws gives the fixed releases as earlier than 24.1.10, 25.0.13 and 26.0.5. The database’s own product configuration data for the same records lists earlier than 24.1.9, 25.0.12 and 26.0.4 [7][8]. The database flags the mismatch and directs readers to the vendor advisory. Anything consuming that configuration data programmatically, including a scanner or an asset inventory rule, will mark builds as remediated that the vendor’s own data says are not. Given that 26.0.4 is the build watchTowr reports as still chainable to unauthenticated code execution [1], the gap between those two lists covers exactly the versions where the question matters.
The Canadian Centre for Cyber Security advisory of 28 August named both CVEs, recorded the KEV additions, and advised installing updates as they became available [6]. That phrase, written before anyone knew how many releases were coming, described the fortnight accurately.
Exploitation ran ahead of the fixes
Activity progressed through recognisable stages. watchTowr’s sensor network captured more than 1,000 exploitation attempts across more than ten geographies, moving from fingerprinting and mass scanning to targeted exploitation and then to hands-on-keyboard activity, and the researchers judged that broader exploitation probably predated their visibility [2]. Huntress observed exploitation in two customer environments and reproduced the full pre-authentication chain, with logged activity consistent with reconnaissance rather than malware deployment or persistence [4]. More than 1,000 instances were internet-exposed according to Shadowserver data cited at the time [10]. The vendor characterised the attacks as limited and targeted [4]. Those figures measure different things and should not be summed.
No actor attribution has been published for this cycle [4][10]. The product’s history is relevant to triage without being predictive: the 2023 PaperCut flaw CVE-2023-27350 was exploited by multiple groups including ransomware operators [5], and the 2023 campaign fell disproportionately on education [12]. PaperCut is widely deployed in higher education, schools, healthcare and enterprise print estates [12], which is a sector profile worth noting in the Asia-Pacific, where university print infrastructure is frequently internet-reachable and rarely owned by the security function.
What to do with this
Four actions follow from the record rather than from the flaws. Establish the exact build number on every PaperCut Application Server, including site servers and secondary print servers, and compare it against the vendor’s version data rather than a scanner verdict derived from database configuration entries [4][7]. Treat any instance that was internet-exposed and unpatched at any point from late August as requiring incident response rather than patching alone [2][10]. Capture forensic evidence, including memory, before restarting the service, because patching does not evict an in-memory implant and a restart destroys it along with the evidence of it [2]. Then remove the exposure: restrict the management interface to trusted addresses by firewall rule, network access control or authenticating reverse proxy, which is the control that would have held through all four bypasses [5][11][12].
Two post-incident checks are easy to miss. External user-lookup database connector settings may not revert to defaults after exploitation, so they should be inspected rather than assumed [2]. And the absence of published network indicators is not evidence that a server is clean, a point the vendor made in its own indicator guidance [3].
The offensive security reading
The useful conclusion is about process, not about print servers. For a fortnight, the honest answer to whether an organisation was exposed depended on which of five releases it held, and that answer could not be derived from the CVE identifier, the severity score, or the KEV deadline. All three were stable while the remediation state underneath them moved four times. Remediation tracking built on CVE status treats patched as a terminal condition. This cycle shows it is a claim with a build number attached, and that the build number is the only part of it worth recording.
Sources
Every R3KONX article cites its primary material. 12 sources, in order of first citation. Links open the original publication.
- Death By A Thousand PaperCuts (PaperCut Pre-Auth RCE Chain and Patch Bypasses WT-2026-0141-0144) watchTowr Labs · 2026-10-09
- PaperCut NG/MF Zero-Day: CVE-2026-81578, CVE-2026-82078 (Active Exploitation Underway) watchTowr · 2026-09-09
- PaperCut warns of NG, MF flaw exploited in zero-day attacks BleepingComputer · 2026-08-27
- PaperCut releases second emergency patch for exploited flaws BleepingComputer · 2026-08-28
- PaperCut NG/MF Critical Zero-Day Exploited in the Wild Rapid7 · 2026-08-28
- PaperCut security advisory (AV26-858) Canadian Centre for Cyber Security · 2026-08-28
- CVE-2026-81578 record National Vulnerability Database, NIST · 2026-08-28
- CVE-2026-82078 record National Vulnerability Database, NIST · 2026-08-28
- CVE-2026-82077 record National Vulnerability Database, NIST · 2026-09-24
- PaperCut Exploitation Escalates to Active Intrusions SecurityWeek · 2026-09-01
- PaperCut Software vulns: CVE-2026-81578, CVE-2026-82078 runZero · 2026-08-27
- CSA Research Note: PaperCut NG/MF Zero-Day, Active Exploitation Cloud Security Alliance · 2026-08-28
Researched and written by the R3KONX analysis desk from the cited primary material. Two caveats apply. The patch and bypass timeline is reconstructed from the researchers' published account and contemporaneous reporting, because the vendor's own bulletin pages did not render retrievable text for this review; dates for individual builds should be verified against the vendor portal before being used for audit. Exploitation scale figures come from sensor networks and incident responders with different visibility and are not comparable. Corrections to event@r3konx.asia.
