Huntress has identified active intrusion activity in which operators are leveraging a cluster of recently disclosed weaknesses in Microsoft Defender to obtain elevated execution context and degrade endpoint protection controls.

Campaign Chains

The campaign chains three publicly released zero-day techniques, BlueHammer, RedSun, and UnDefend, that originally published by the researcher operating under the aliases Chaotic Eclipse and Nightmare-Eclipse following a dispute over disclosure timelines. The researcher expressed frustration with Microsoft’s handling of the vulnerability disclosure process on April 2, 2026, stating they had not wanted to release new exploit code but felt compelled to do so after a breakdown in the agreed-upon terms.From a technical standpoint, both BlueHammer and RedSun represent local privilege escalation primitives targeting Defender’s service architecture and its interaction with core Windows filesystem and security boundaries. These flaws are exploitable post-compromise, meaning an adversary must already possess code execution, typically as a low-privileged user, before invoking the exploit chain. Once triggered, the vulnerabilities allow escalation to NT AUTHORITY\SYSTEM, effectively bypassing standard user-mode restrictions and granting full control over the host, including the ability to disable security tools, tamper with logging, and deploy further malware before moving laterally.

1. BlueHammer

BlueHammer, now tracked as CVE-2026-33825, is rooted in a time-of-check to time-of-use (TOCTOU) race condition within Windows Defender’s threat remediation engine. The vulnerability exists because Defender performs privileged file operations during malware cleanup without adequately validating the file path at the time of the actual write operation, creating a window during which an attacker can redirect that operation through filesystem manipulation. The exploit triggers a Defender detection and then uses a batch opportunistic lock (oplock) to pause Defender’s file operation at the precise moment between the check and the write. During this pause, the attacker constructs an NTFS junction point, a filesystem-level directory symlink, that redirects Defender’s target path from an attacker-controlled temporary directory to C:\Windows\System32. When the oplock is released and Defender resumes the write, it follows the junction and overwrites a system binary under its own SYSTEM-level privileges, delivering arbitrary code execution in the most privileged local context on the machine. The flaw is classified under CWE-1220 (insufficient granularity of access control), requires no user interaction, and carries a CVSS score of 7.8 (High). Microsoft addressed BlueHammer through Defender Antimalware Platform update version 4.18.26030.3011, delivered as part of the April 2026 Patch Tuesday cycle.

2. RedSun

RedSun, while sharing a surface family resemblance to BlueHammer, exploits an entirely distinct and as-yet unpatched code path within Defender’s cloud file handling subsystem. At the heart of the vulnerability is a logic flaw in how Defender processes cloud-tagged files: when it detects a malicious file bearing a cloud tag, rather than simply quarantining or deleting it, Defender attempts to rewrite the file back to its original location, a behavior that RedSun weaponizes through a carefully constructed attack chain. The exploit registers a Cloud Files sync root via CfRegisterSyncRoot(), using the same API employed by OneDrive and Dropbox, and then drops a Cloud Files placeholder via CfCreatePlaceholders() carrying extended attributes that mark the file as a remote-backed file pending hydration. When Defender detects this cloud-tagged file and initiates its rollback, the attacker uses an oplock to pause the restoration mid-operation, then substitutes an NTFS junction that redirects the target path to a privileged directory. The end result is that Defender writes an attacker-controlled binary to C:\Windows\System32\TieringEngineService.exe, after which the Cloud Files Infrastructure service executes that binary as SYSTEM. Critically, the technique requires no memory corruption, no kernel bug, no code injection primitive, and no internet connection so it chains entirely legitimate Windows features. Researchers have documented RedSun functioning with near-100% reliability on fully patched Windows 10, Windows 11, and Windows Server 2019 and later systems where cldapi.dll is present. No CVE has been assigned and no patch exists as of April 18, 2026.

3. UnDefend

The third component, UnDefend diverges from privilege escalation and instead targets defensive impairment through a denial-of-service condition against Defender’s signature update mechanism. Together with BlueHammer and RedSun, it completes a three-part offensive toolkit: one technique enables privilege escalation via a privileged file read path, another enables escalation via a privileged file write path, and the third gradually blinds Defender to new threats by starving it of updated definitions. The operational consequence of deploying UnDefend is cumulative: as the endpoint falls further behind on threat intelligence, adversaries operating in that environment face progressively diminishing detection risk, effectively extending the window during which they can conduct reconnaissance, move laterally, or stage payloads without triggering signature-based controls.

Threat Actors in Action

Huntress telemetry shows that exploitation follows a recognizable hands-on-keyboard pattern consistent with post-exploitation frameworks. Observed enumeration commands include whoami /priv to assess token privileges, cmdkey /list to surface stored credentials, and net group to map domain or local group memberships which are a sequence that indicates direct operator involvement rather than fully automated malware deployment. Initial access in at least some of the observed incidents was obtained through compromised SSL-VPN credentials, with the Defender exploit chain applied as a post-access privilege escalation mechanism rather than as an initial entry point.

Threat actors began actively pairing RedSun with BlueHammer and UnDefend in real incidents as of April 16, 2026, with observed tactics including staging renamed exploit binaries in user-writable directories such as Pictures or Downloads, applying the chain to gain SYSTEM after VPN-based initial access, and combining RedSun with UnDefend to suppress Defender updates and consolidate persistence. From a detection engineering standpoint, the most reliable host-based indicators include unexpected modification of C:\Windows\System32\TieringEngineService.exe (the direct write target of RedSun), Defender events involving cloud-tagged files or EICAR-like detections followed immediately by file restoration activity, and the creation of NTFS junctions or reparse points in close temporal proximity to Defender remediation events.

Microsoft has confirmed that BlueHammer has been remediated through the patch associated with CVE-2026-33825, though the company made no mention of public exploit code in its advisory. RedSun and UnDefend remain without official fixes at the time of writing. This leaves a residual risk window that is particularly acute given that the common architectural factor across all three techniques is Defender itself — a privileged, always-on process that must interact with every file the user touches that is suggesting the weaknesses identified are systemic rather than incidental.

What to Do

From a defensive engineering perspective, this cluster highlights a critical and underappreciated risk category: exploitation of security tooling as an attack surface in its own right. Mitigation strategies should prioritize verification that Defender has received the April 2026 Antimalware Platform update, monitoring for anomalous MsMpEng.exe behavior particularly around file write operations involving cldapi.dll, baselining and integrity-monitoring the SHA-256 hash of TieringEngineService.exe across endpoints, and deploying complementary controls such as EDR telemetry correlation and kernel integrity monitoring to surface abnormal privilege transitions and service tampering that bypass signature-based detection entirely.

References: