Week of June 23, 2025

Microsoft announced a preview of new Windows endpoint security platform capabilities designed to enable security vendors to build solutions that operate outside the kernel. This architectural shift aims to improve system stability and security by reducing reliance on kernel-mode drivers. The move could reshape how endpoint detection and response products interact with the Windows OS.

A proof-of-concept tool named ValorantBypass was published on GitHub, demonstrating how to temporarily disable and suspend Vanguard anti-cheat processes to bypass the TPM/Secure Boot requirement popup in Valorant. The tool stops the Vanguard service, suspends the watchdog process, and suspends svchost.exe instances loading tpmcore.dll to suppress the popup. It is intended for educational and security research purposes.

Check Point Research discovered the first documented case of malware embedding prompt injection to evade AI-based detection. The technique manipulates AI models into misclassifying malicious content as benign. This represents a novel evasion method targeting AI-powered security tools.

A research article details multiple techniques to execute Metasploit Meterpreter payloads on a fully patched Windows 11 system with Defender enabled. The methods include remote mapping injection, AMSI bypass, sandbox evasion via delayed execution, PPID spoofing, and Defender exclusion abuse. These techniques exploit Windows API quirks rather than vulnerabilities, highlighting the need for layered endpoint defenses.

A critical OS command injection vulnerability (CVE-2025-34041) exists in the Chinese-language versions of Sangfor Endpoint Detection and Response (EDR) management platform versions 3.2.16, 3.2.17, and 3.2.19. Unauthenticated attackers can send malicious HTTP requests to the EDR Manager interface to achieve arbitrary command execution with elevated privileges. Exploitation was observed in the wild by the Shadowserver Foundation on 2025-02-04.

A proof-of-concept tool named TrampoLatte demonstrates bypassing AMSI and ETW by using trampoline hooks to alter function execution. It hooks AmsiScanBuffer to always return AMSI_RESULT_CLEAN and hooks EtwpEventWriteFull to return immediately, preventing detection. This technique could allow malicious code to evade endpoint security monitoring.

A security researcher demonstrated a technique to bypass Carbon Black's application whitelisting and Cortex XDR's detection by abusing InstallUtil.exe to execute arbitrary C# code, leading to process injection and a reverse shell on a hardened Windows 10 host. The method leverages the Uninstall() function in a crafted .NET executable to perform code injection without triggering endpoint defenses. This highlights a gap in monitoring for legitimate signed binaries used for living-off-the-land attacks.

A vulnerability in Insyde H2O UEFI firmware allows an attacker with local admin access to bypass Secure Boot and modify the DXE volume during firmware updates. By shadowing NVRAM variables, an attacker can make the firmware trust a self-signed certificate, enabling execution of arbitrary code in the firmware update context where write protections are removed. This can lead to persistent firmware compromise if the FlashDeviceMap is misconfigured.