-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256 NEFARIOUSPLAN-CANONICAL-V1 {"body_md":"> The CVE record names the patched surface. The patched surface is one of several.\n\nFrom my file on [CVE-2026-26128: Microsoft Titled It SMB. The Primitive Is DnsCache and the KDC Disagreeing About a Hostname.](/posts/cve-2026-26128-smb-was-not-the-bug), the March 2026 Windows SMB Server advisory whose actual primitive was two Windows services calling two different Unicode normalization functions on the same hostname.\n\nThose two sentences frame every other case file on this site. Read the fourteen on Microsoft in a row and the shape is this: Microsoft repeatedly ships two or more of its own components that encode different answers to a shared question, writes an advisory locating the bug inside one of them, and patches that one. The advisory is a scope. The bug is a gap.\n\n## The audit tool and the mitigation tool shipped from the same directory\n\nMy file on [CVE-2026-42897](/posts/exchange-cve-2026-42897-mitigation-invisible-to-health-checker) opens with two Microsoft modules in one GitHub repository, `microsoft/CSS-Exchange`. One is the Exchange On-premises Mitigation Tool, Microsoft's emergency stopgap that installs IIS configuration when a critical bug needs to be neutralized before a Cumulative Update can ship. For CVE-2026-42897, it installs a Content-Security-Policy rewrite rule named `EOMT OWA CSP - outbound` in the only section of the IIS configuration where a response header can live, ``. The other module is Exchange Health Checker, the diagnostic an administrator runs to confirm the first one did its job. Its enumeration function, `Get-URLRewriteRule.ps1`, reads IIS rewrite configuration from three sources.\n\n> Three reads of `.rewrite.rules`. Zero reads of `.rewrite.outboundRules`.\n\nBoth engineers were correct in isolation. The mitigation team installed the header where an IIS response header can live. The audit team wrote for the common case, inbound rewrite rules. Nobody reconciled. The server that needs the mitigation and the server that already has it produce the same Health Checker report, the one with `EOMT OWA CSP - outbound: NOT DETECTED` on both lines.\n\n## A pointer stashed in two structures, freed by both destructors\n\n[CVE-2026-33824: Windows IKE allocated a blob at MMSA+0x208, reassembly shallow-copied the pointer into PacketContext+0xC8, and both destructors ran](/posts/windows-ike-cve-2026-33824-two-owners-one-free) is this disagreement at the C language level. `IkeHandleSecurityRealmVendorId`, a kernel-mode handler in `ikeext.sys`, allocates a vendor-ID blob on the paged pool and stores the pointer at `MMSA+0x208` on the Main-Mode Security Association structure. When a fragmented `IKE_AUTH` arrives, `IkeReinjectReassembledPacket` shallow-copies that pointer into `PacketContext+0xC8` on a fresh packet-context structure. The malformed-packet error path then tears down both.\n\n> One line copied a pointer that should have been moved. Two destructors, each correct, freed the same heap address.\n\n`IkeDestroyPacketContext` and `IkeFreeMMSA` each free the thing they are told they own. Three pre-authentication remote code executions in `ikeext.sys` since September 2022 (CVE-2022-34721, CVE-2022-34722, CVE-2026-33824), same driver, same CVSS 9.8, same shape. The ownership convention lives in the programmer's head. The programmer rotates.\n\n## The handler for class 253 reused a different local variable\n\n[CVE-2026-40369: ntoskrnl Set STATUS_INFO_LENGTH_MISMATCH And Did The Writes Anyway](/posts/ntoskrnl-cve-2026-40369-mismatch-then-writes) is `ProbeForWrite`, the Windows kernel's user-pointer validator, and `ExpGetProcessInformation`, its caller, disagreeing about what the sizing convention of `NtQuerySystemInformation` means. `ProbeForWrite` with `Length=0` returns immediately, because there is nothing to probe. The handler for `SystemInformationClass` 253 computes `STATUS_INFO_LENGTH_MISMATCH`, stores it in a local named `status`, and keeps going, writing three DWORDs at the caller-supplied kernel address for every process in the system.\n\n> The error code goes into a local variable named `status`. The conditional that gates the writes is `infoClass == 253`, not `status == 0`.\n\nClasses 5 and 252, the older arms of the same dispatcher, NULL-check their local pointer before dereferencing. Class 253 was written later, reused a fresh local, and inherited none of the precautions. Five implementations of \"read per-process accounting into a caller-named buffer\" sit inside one function body. Four of them check. One does not.\n\n## The firmware checks the signer, the patch replaces the file\n\n[CVE-2025-48804: The Pre-Patch bootmgfw.efi Is Still Signed](/posts/cve-2025-48804-pre-patch-bootmgr-still-signed) runs this disagreement under the operating system. Microsoft patched a flaw in the Windows boot manager's handling of System Deployment Image files in July 2025, shipping a new `bootmgfw.efi`. The pre-patch binary chains to the same issuer, *Microsoft Windows Production PCA 2011*, as the replacement. Secure Boot validates the issuer.\n\n> The patched bootmgfw.efi supersedes the pre-patch one in every way except the one Secure Boot checks.\n\nMicrosoft has the mechanism to retire the 2011 trust anchor. KB5025885, a multi-stage opt-in DBX rollout that walks administrators through refusing binaries signed by the PCA 2011, published in May 2023. Microsoft has not driven it by default in the three calendar years since. The patch and the revocation live at different layers. Microsoft owns both. One shipped as a monthly cumulative. The other has been an administrator checkbox for thirty-six months.\n\n## The helper function hands out the precondition\n\n[BlueHammer: What the Researcher Commented Out](/posts/bluhammer) and [CVE-2025-21333: The Defender Hands Out the VmId](/posts/cve-2025-21333-the-defender-hands-out-the-vmid) are the Microsoft Defender instance of the pattern. BlueHammer's patch closed the specific definition-update path a researcher used to redirect a SYSTEM-level Defender write through the NT object namespace.\n\n> Microsoft fixed the definition update path. The IMpService RPC interface that triggered the vulnerable operation is undocumented and has other callable methods.\n\nThe RPC interface is unchanged. For CVE-2025-21333, Microsoft patched an integer-truncation overflow in `vkrnlintvsp.sys`, the Hyper-V Kernel Integration VSP driver. The second public exploit skipped spawning Windows Sandbox entirely and called `CoCreateInstance` on `CLSID_IsolatedAppLauncher`, the documented COM entry point Microsoft ships to support Microsoft Defender Application Guard, to retrieve the container GUID the kernel bug required. The advisory named `vkrnlintvsp.sys`. The research named the signed COM service next to it.\n\n## The subsystem has at least four answers to one question\n\n[CVE-2026-26128](/posts/cve-2026-26128-smb-was-not-the-bug), the file the opening blockquote comes from, is the purest case. Microsoft titled the advisory *Windows SMB Server Elevation of Privilege*. The primitive is two normalization functions in two services inside the same `lsass.exe` process tree, answering \"are these two hostnames the same string?\" with different flag bags. DnsCache calls `CompareStringW(NORM_IGNORECASE)` on a hostname with one circled letter in it and returns not-equal, letting the resolution proceed to the network. The Key Distribution Center calls `LCMapStringEx(0x31403)` with `NORM_IGNOREWIDTH` set on the same hostname and returns the same sort key as the real Domain Controller, binding the ticket to its own machine account.\n\nThe March 2026 patch closed the SMB relay target the advisory text named. The PoC's README lists ADCS web enrollment and MSSQL as unpatched relay targets. The authentication subsystem has no canonical answer to \"is this hostname us?\"; it has at least four, each chosen for the team that wrote it.\n\n## The shell carries the same disagreement one layer up\n\nThree shell-handler advisories in five months close three render paths while the handler underneath remains. My file on [CVE-2025-33053](/posts/cve-2025-33053-the-other-half-got-a-doc-edit) patched the `WorkingDirectory=` directive in `.URL` files, the half that let Stealth Falcon redirect a LOLBin's working directory onto an attacker WebDAV share. The `URL=` directive, which does the same thing more directly, received a wording change on the Microsoft Learn page instead of a CVE. [CVE-2025-24054: The Two Verbs That Don't Trigger Are 'Open' and 'Execute'](/posts/cve-2025-24054-open-and-execute-dont-trigger) patched one selection trigger in `.library-ms` files; opening and executing the file are the only two interactions a Windows user can have with it that did not fire the hash leak. My file on [CVE-2026-32202](/posts/cve-2026-32202-trust-check-at-the-click) patched Control Panel applet IDList trust at `ShellExecuteExW`, the API that runs when the user double-clicks the shortcut; icon extraction had already opened an SMB session to the attacker's server by the time the double-click was available. Preview pane, tooltip resolution, infotip composition, and property-handler walk have each been named across these files as render paths that touch attacker-supplied strings without consulting the trust pipeline. The patches land at individual entry points. The format exposes the primitive by design.\n\n## The last call landed\n\nThe previous plan predicted that the next in-the-wild exploited Microsoft advisory in this pattern would carry a title naming one Windows component, a research writeup identifying at least two components the title did not own, and a patch landing in only the named component, with the specific call being an ADCS-reachable Kerberos relay path whose advisory did not mention ADCS. CVE-2026-26128 is that advisory. The title is *Windows SMB Server Elevation of Privilege*. The research names DnsCache, the KDC, `CompareStringW`, and `LCMapStringEx`. The patch closes the SMB relay path. The PoC's README states that ADCS web enrollment is a supported, unpatched relay target. Called.\n\n## Where the files acquit Microsoft\n\nNot everything in the catalog is unreconciled components. [CVE-2025-50165: The JPEG Is the Decoy. The PIF Module Is the Payload.](/posts/cve-2025-50165-pif-module-is-the-payload) and [CVE-2025-47987: The CredSSP PoC Documents A Project That Does Not Exist](/posts/credssp-cve-2025-47987-poc-documents-a-different-project) are about placeholder PoCs on GitHub, assembled by strangers who never ran what they committed. That is what the ecosystem around Microsoft's CVE numbers has become. It is not Microsoft's plan. [CVE-2010-0249: The KEV Deadline for Aurora Is a Census](/posts/cve-2010-0249-kev-deadline-is-a-census) is sixteen years old; MS10-002 was Microsoft's first modern out-of-band emergency patch, shipped nine days after Google disclosed Operation Aurora, and the 2026 KEV deadline is a federal inventory event about a retired product, not a current patching failure. [CLFS: Ransomware's Favorite Kernel Driver](/posts/clfs-ransomwares-favorite-kernel-driver) is the adjacent charge, five exploited-in-wild use-after-frees in `clfs.sys` in three years, each closed at instance scope. One team, one object lifecycle model whose edges keep producing the same bug. The charge there is about the scope of the patch, but the mechanism is different: there are not two components disagreeing, there is one team reusing a model with no conclusion drawn. The patches themselves do what the advisories say. The CVE-2026-33824 fix closed the exact double-free the PoC fired. The `WorkingDirectory=` half of CVE-2025-33053 closed the LOLBin chain Stealth Falcon used in the wild. The charge is not that the patches fail. The charge is the scope they are drawn to.\n\n## Where to look first\n\n`ikeext.sys` is the most productive single driver for this pattern. Three pre-authentication RCEs since September 2022, the ownership convention around attacker-controlled blob pointers unchanged across all three. Any function in that driver that stashes a pointer to attacker-influenced bytes on a long-lived state object, and whose pointer later gets shallow-copied onto a shorter-lived one, is the place.\n\nThe shell handlers for file formats whose body embeds a URL are the second place. The `.URL`, `.library-ms`, and Control Panel applet IDList in `.lnk` are each called from icon extraction, from tooltip composition, from preview-pane thumbnail providers, and from the property-handler walk. Three of those four render paths have produced advisories in this catalog. The fourth has not.\n\n`CLSID_IsolatedAppLauncher`'s remaining methods are the third. The interface's IDL is documented in the Windows SDK. One method hands out a Hyper-V container GUID to any process in the session, with no privilege gate. The other methods on that interface have not been audited for callers that hold neither privilege nor a legitimate isolation session.\n\n## The next advisory\n\nThe next in-the-wild exploited Microsoft advisory in this pattern will be a pre-authentication remote code execution in `ikeext.sys`, whose advisory names one function in the driver and whose patch lands in only that function. The alternative is a sixth use-after-free local privilege escalation in `clfs.sys`, exploited in the wild before the patch, attributed to a financially-motivated ransomware operator.","closing_line":"You will patch a function in `ikeext.sys`, Microsoft. The function on the other side of the handoff is also yours.","hook_md":"Across fourteen case files, the Microsoft advisory names one component and the research names at least two. The patch lands inside the one. The disagreement that produced the bug ships again, next driver, next quarter, next advisory.","post_id":885,"slug":"the-nefarious-plan-of-microsoft-2","title":"The Nefarious Plan of Microsoft","type":"editorial","unreadable_sentence":"The ownership convention lives in the programmer's head."} -----BEGIN PGP SIGNATURE----- iHUEARYIAB0WIQRf0htP5+SjynlxywneZjl4jgkQJgUCawHrHAAKCRDeZjl4jgkQ JgasAQC4XCl/5CSwcnrNJGucHn3w0+xxq75sdwG+ZuqdgDg/+wEA3gBUTntGM+ym crMC+7c2CsDgtzuvZe/AVx1Ei0RHkgg= =dC1N -----END PGP SIGNATURE-----