Week of May 11, 2026

On July 19, 2024, a faulty Rapid Response Content configuration file (Channel File 291) for the CrowdStrike Falcon sensor caused widespread Windows Blue Screen of Death (BSOD) crashes. The update, deployed without staged rollout, affected approximately 8.5 million devices globally, disrupting airlines, hospitals, and critical services. The incident stemmed from an architectural decision to allow rapid threat-detection updates to the kernel-level sensor without the same testing rigor applied to sensor code.

A security researcher used AI to iteratively develop a PowerShell script that bypasses the Antimalware Scan Interface (AMSI) by patching the AmsiScanBuffer function in memory. The AI-assisted workflow demonstrates how attackers can leverage large language models to refine evasion techniques against endpoint detection mechanisms.

Security researcher Vladimir Tokarev discovered four vulnerabilities in OpenClaw that allow data theft, privilege escalation, and persistence. The flaws can be exploited by weaponizing the agent's own privileges. Users are advised to update to the latest version to mitigate the risks.

A new variant of the Gremlin stealer malware has been identified using advanced obfuscation to conceal its command-and-control infrastructure and data exfiltration paths. The malware hides C2 details and exfiltration routes within encrypted resources, making detection and analysis more difficult. This technique poses a challenge for endpoint security solutions that rely on network traffic analysis or static signatures.

A local privilege escalation vulnerability exists in Rapid7 Metasploit Pro on Windows. The metasploitPostgreSQL service runs as SYSTEM and loads an OpenSSL configuration file from a static, world-writable location. A pre-existing unprivileged 'vagrant' user can plant a malicious openssl.cnf to execute arbitrary commands with SYSTEM privileges, leading to full host compromise.

OpenAI confirmed that two employee devices were compromised in the TanStack supply chain attack, which affected hundreds of npm and PyPI packages. The breach prompted OpenAI to rotate code-signing certificates for its applications as a precaution.

OpenAI announced Daybreak, an AI-driven vulnerability discovery initiative, in partnership with Cisco and CrowdStrike. The project aims to use frontier models to find vulnerabilities in products and share open-source frameworks to improve security across the industry.

ESET discovered new FrostyNeighbor activity since March 2026 targeting Ukrainian governmental organizations with spearphishing PDFs containing malicious links. The campaign uses a JavaScript variant of PicassoLoader to deliver Cobalt Strike beacons, with server-side victim validation to evade detection. This shows continued evolution of the group's espionage toolset.

The Iran-linked Seedworm (MuddyWater) APT group conducted a global espionage campaign in early 2026, targeting at least nine organizations across multiple sectors. The attackers used DLL sideloading with a legitimate signed SentinelOne binary (sentinelmemoryscanner.exe) to load a malicious DLL (sentinelagentcore.dll), evading endpoint detection. The malicious DLL deployed ChromElevator to steal browser credentials, with the attack chain driven by a Node.js script.

Microsoft has released patches for two zero-day vulnerabilities in Microsoft Defender, named UnDefend and RedSun, which were publicly disclosed last month. The flaws were actively exploited in the wild before the patch was available.

Microsoft patched multiple vulnerabilities in its Defender security product, addressing flaws that could allow attackers to bypass detection or elevate privileges. Separately, a BitLocker bypass technique was disclosed, potentially enabling attackers to access encrypted data under certain conditions.

Microsoft acknowledged a security feature bypass vulnerability in Windows tracked as CVE-2026-45585 and publicly referred to as 'YellowKey'. The flaw is being exploited in the wild, and Microsoft has shared mitigation measures to defend against potential attacks.

A disgruntled researcher publicly released two zero-day vulnerabilities affecting Microsoft products. One flaw allows credential dumping from the LSASS process while bypassing Microsoft Defender for Endpoint detection. The disclosure raises concerns about increased risk of credential theft on compromised endpoints.

In Q1 2026, Lyrie observed a new ransomware family, MORPHIC, that dynamically alters its encryption speed, pauses operations, switches encryption algorithms, and rewrites in-memory code in response to detected security tools. This runtime adaptation defeated traditional behavioral detection and static honeypots across four customer environments.

Palo Alto Networks disclosed multiple authorization bypass vulnerabilities in the Endpoint DLP component of Prisma Access Agent. A local attacker with low privileges can bypass authentication to execute privileged operations, potentially compromising confidentiality, integrity, and availability of the endpoint. The issue affects versions prior to 26.2.1 on macOS and Windows.

The Nitrogen ransomware group breached Foxconn's North American manufacturing facilities in May 2026, exfiltrating 8TB of data including hardware schematics for major tech customers. The attack used malvertising and DLL sideloading to gain initial access, followed by weeks of lateral movement and data staging before deploying ransomware. This incident highlights the supply chain risk posed by ransomware groups targeting manufacturing to steal sensitive intellectual property.

A red team engagement demonstrated a five-step attack chain against Microsoft Entra ID, starting from a single set of valid credentials blocked by Conditional Access. By exploiting gaps in Device Registration Service, phantom device registration, and Intune compliance evaluation, attackers achieved Global Administrator access without touching a corporate endpoint or deploying malware. The technique has been observed in the wild by the Storm-2372 APT group.

SentinelOne has updated its Windows agent installer as part of the Guardian installer / SentinelInstallerDriver security hardening initiative. The change improves protection around installer and upgrade workflows and aims to reduce exposure to Bring Your Own Installer (BYOI) bypass attacks.

CISA added CVE-2023-36010, a denial-of-service vulnerability in Microsoft Defender, to its Known Exploited Vulnerabilities catalog, indicating active exploitation in the wild. The flaw allows an attacker to cause a denial of service on affected systems. This highlights the risk of vulnerabilities within endpoint security agents themselves.

Security researchers demonstrated YellowKey, a zero-day exploit that decrypts Microsoft BitLocker-protected drives by placing specific files on a USB stick, bypassing the need for the recovery key or password. The attack requires physical access and leverages an apparent backdoor in the BitLocker implementation. A second zero-day, GreenPlasma, enables local privilege escalation.

Microsoft observed a stealthy intrusion campaign where attackers bypassed traditional malware by abusing the legitimate HPE Operations Agent. The technique allowed silent network infiltration without dropping malicious files, evading endpoint detection. This highlights the risk of trusted enterprise tools being weaponized for lateral movement and persistence.

A new Vidar Stealer campaign targets Windows endpoints, employing stealthy methods to evade endpoint detection and response (EDR) systems. The malware steals passwords, browser cookies, cryptocurrency wallets, and authentication data. This campaign highlights evolving evasion tactics that challenge current endpoint security controls.

A method is described to identify which functions in ntdll.dll are hooked by EDR/AV products by comparing the first bytes of exported functions in memory against a clean copy from disk. This reveals monitored vs. unmonitored syscall paths, enabling attackers to choose evasion routes. The technique filters out data exports to reduce false positives and classifies hook types.

A vulnerability in Wellbia's XIGNCODE3 anti-cheat driver xhunter1.sys allows an unprivileged user-mode process to obtain a PROCESS_ALL_ACCESS handle to any process, including Protected Process Light (PPL) targets like lsass.exe. The driver's IRP_MJ_WRITE handler for command 785 calls ObOpenObjectByPointer with KernelMode access mode, bypassing access checks, and omits OBJ_KERNEL_HANDLE, placing the handle in the caller's process handle table. This enables full read/write/terminate access, defeating PPL protections on fully-patched Windows 11 systems.

Attackers are distributing fake Claude Code installers that deploy malware to steal protected browser data. The malware abuses Chrome's IElevator interface to escalate privileges and access encrypted credentials and cookies. This technique bypasses standard browser protections, enabling theft of sensitive authentication material.

In April 2026, Vercel disclosed a breach where an attacker used OAuth tokens stolen from Context.ai via Lumma Stealer malware to access Vercel internal systems and customer environment variables. The attack chain started with a Context.ai employee downloading Roblox auto-farm scripts, leading to a two-month dwell time and eventual exposure of credentials for AWS, Azure, GCP, GitHub, Stripe, Twilio, and SendGrid. The incident highlights risks of OAuth trust relationships and AI-supply-chain attacks.

CVE-2026-42869 is an authentication bypass vulnerability in SOCFortress CoPilot versions prior to 0.1.57. The platform ships with a hardcoded JWT signing secret, allowing unauthenticated attackers to forge admin tokens and gain full control of CoPilot and connected security tools. The flaw arises from a fallback secret in the source code and default Docker Compose configuration.

Google Threat Intelligence Group (GTIG) observed a zero-day exploit assessed to have been developed with AI assistance. The report also notes AI-generated decoy code in malware campaigns. This marks a notable instance of AI being used to create novel exploits.

An intrusion began with a malicious MSI masquerading as Sysinternals RAMMap, deploying an EtherRAT variant that used Ethereum blockchain for C2 configuration. The threat actor later deployed TukTuk malware, a new AI-generated framework using SaaS platforms for C2, and ultimately executed The Gentleman ransomware across the domain via GPO after disabling defenses and exfiltrating data.

Research demonstrates that AI can analyze software patches and generate working exploits in as little as 30 minutes, drastically reducing the time between patch release and exploit availability. This challenges the traditional 90-day disclosure window and demands faster patch deployment and revised disclosure practices.

Google's research indicates cybercriminals and nation-state actors are using AI for vulnerability research, exploit development, and creating malware with decoy logic to evade endpoint detection. AI-generated media like voice cloning is also used to bypass two-factor authentication. This highlights an emerging trend of AI lowering the barrier for sophisticated endpoint attacks.

GhostLock is a proof-of-concept tool that locks file access on Windows using legitimate API calls, achieving ransomware-like business disruption without encrypting files. It demonstrates a novel availability attack that bypasses traditional detection mechanisms focused on encryption behavior.