Week of May 18, 2026

A supply chain attack dubbed 'TrapDoor' uses hidden code in AI assistant configurations to steal cryptocurrency wallets and cloud credentials from blockchain engineers. The malware specifically targets developer endpoints, compromising sensitive data through malicious AI tool integrations.

Financially motivated threat actors are using SEO poisoning to trick developers into downloading fake installers for Google Gemini CLI and Anthropic Claude Code. The campaign delivers a fileless PowerShell infostealer that targets developer workstations, stealing credentials and sensitive data.

CVE-2013-3900 is a vulnerability in the Windows WinVerifyTrust function that validates Authenticode signatures on portable executable files. Attackers can append malicious data to a signed file without invalidating the signature, potentially allowing unsigned code to execute while appearing trusted. This undermines endpoint security mechanisms that rely on code-signing verification.

Google inadvertently published exploit code for a high-severity Chromium sandbox escape vulnerability (CVE-2025-2783) before a patch was available. The flaw allows attackers to bypass the browser's sandbox protections, potentially leading to arbitrary code execution. The premature disclosure increases the risk of in-the-wild exploitation before users can apply the fix.

A threat actor compromised an end-of-life F5 BIG-IP edge appliance and pivoted to an internal Linux host via SSH. From there, they exploited an unpatched Atlassian Confluence server to gain code execution and used relay-style authentication attacks against Active Directory. The incident highlights risks from unpatched edge devices and SaaS applications enabling lateral movement to identity systems.

A Google study reveals that LLMs can generate malware that evades detection by dynamically changing its code structure and behavior. This morphing capability allows the malware to bypass signature-based and heuristic endpoint security tools. The research highlights an emerging threat where AI-generated attacks adapt faster than traditional defenses.

Fox-IT reports on RemotePE, a remote access trojan used by a Lazarus subgroup in memory-only attacks against financial and cryptocurrency organizations. The malware operates entirely in memory, leaving minimal forensic artifacts and evading traditional file-based detection.

Google's Project Zero developed a zero-click exploit chain targeting Pixel 10 phones, achieving full escalation from remote access to kernel-level compromise without user interaction. The research highlights significant security weaknesses in the device's attack surface and serves as a proof-of-concept to drive improvements in mobile endpoint security.

Unit 42 reports that Iranian APT group Screening Serpens is conducting espionage campaigns targeting technology and defense sectors. The group leverages AppDomainManager hijacking for persistence and deploys new remote access trojan (RAT) variants to maintain access and exfiltrate data.

Researchers demonstrate that software-emulated device nodes can bypass hardware gating in Windows, allowing exploitation of vulnerable drivers from userland without physical hardware. This expands the Bring Your Own Vulnerable Driver (BYOVD) attack surface by making previously hardware-dependent driver bugs exploitable in software-only scenarios.

Attackers are leveraging Brazil's electronic invoice system (NF-e) as a lure to deliver Banana RAT, a banking trojan. The campaign uses social engineering to trick users into executing the malware, which then establishes persistence and steals financial information. This incident highlights the use of regionalized lures to bypass user suspicion and endpoint defenses.

The Cloud Atlas threat group conducted phishing campaigns targeting Russian and Belarusian organizations using malicious LNK files in ZIP archives to execute PowerShell scripts. The scripts deployed VBCloud for file exfiltration and PowerShower for network reconnaissance and lateral movement, employing techniques like UAC bypass and Kerberoasting. The campaign demonstrates evolving tooling and persistence mechanisms.

Splunk released patches for three vulnerabilities: CVE-2025-33883 (information disclosure), CVE-2025-33884 (denial of service), and CVE-2025-33885 (denial of service). These flaws allow low-privileged users to access sensitive data or cause service disruptions. The vulnerabilities affect Splunk Enterprise and Splunk Cloud Platform.

A spear-phishing campaign was observed using nested macOS-style folder structures to hide malware and evade detection by security scans. The technique abuses the way some security tools parse or trust certain file system artifacts, allowing malicious payloads to remain undetected. This highlights a gap in endpoint detection mechanisms when handling non-standard folder formats.

A researcher demonstrated a method to disable the Antimalware Scan Interface (AMSI) in Windows Scripting Host (WSH) without admin privileges or registry changes. By copying WSCRIPT.EXE to AMSI.DLL and executing it, the scripting engine loads the renamed executable instead of the real AMSI.DLL, causing AMSI initialization to fail silently while scripts continue to run uninspected.

Microsoft disclosed and patched a zero-day privilege escalation vulnerability, CVE-2026-41091, in the Microsoft Malware Protection Engine (mpengine.dll). The flaw affects multiple endpoint security products including Microsoft Defender and System Center Endpoint Protection. It was exploited in the wild before a patch was available.

Attackers are distributing a sophisticated variant of the ValleyRAT malware by tricking users into downloading it from fake Microsoft Teams download pages. The malware provides remote access and control over compromised endpoints, posing a significant threat to endpoint security. This campaign highlights the use of social engineering to bypass initial security layers.

Huntress investigated two incidents in April and May 2026 involving The Gentlemen ransomware-as-a-service operation. In both cases, attackers cleared Security, System, and Application Event Logs and used PowerShell to disable Microsoft Defender and add antivirus exclusions. A leak of the operation's internal chats revealed additional evasion tools and techniques.

A threat actor compromised an @antv maintainer account and published malicious versions of widely used npm data-visualization packages. The malicious payload executes during npm install, targeting GitHub Actions environments to steal credentials from multiple platforms. The attack propagated through dependency chains, affecting downstream packages with millions of weekly downloads.

CVE-2026-20206 is a command injection vulnerability in the BrowserBot component of Cisco ThousandEyes Enterprise Agent. An authenticated attacker with transaction test management privileges can execute arbitrary OS commands inside the BrowserBot container. Cisco has addressed the issue in the SaaS service, requiring no customer action.

Attackers compromised a GitHub employee's device using a weaponized Visual Studio Code extension, leading to unauthorized access and exfiltration of data from internal source code repositories. The breach was confirmed by GitHub on May 18, 2026.

A campaign tracked as TAX#TRIDENT is actively targeting Windows users with fake Indian Income Tax assessment pages to distribute malware. The attack uses social engineering to trick victims into downloading and executing malicious files. This incident highlights the ongoing risk of user-targeted malware delivery that can bypass endpoint defenses if users are deceived.

A malware campaign named TamperedChef distributes trojanized productivity applications (PDF editors, calendar tools, file converters) signed with valid certificates to evade detection. The malware delivers information stealers and remote access trojans (RATs), posing a threat to endpoint security by bypassing trust-based checks.

ESET researchers detail the 2025 activities of the China-aligned Webworm APT group, which shifted targeting to Europe and added new backdoors EchoCreep and GraphWorm that abuse Discord and Microsoft Graph API for command and control. The group also employs custom proxy tools and stages malware on GitHub, enhancing stealth and evasion.

The Fox Tempest malware-signing service exploited Microsoft's Artifact Signing infrastructure to digitally sign malicious code. This allowed attackers to bypass endpoint security defenses that trust Microsoft-signed binaries. The incident highlights a novel abuse of trusted code-signing mechanisms to evade detection.

FORTiSEC researchers developed 'Micro-Stager', a technique to evade ETW Threat Intelligence telemetry by leveraging the NtContinue API to set hardware breakpoints without triggering kernel sensors. The method bypasses the CLR crash that occurs when calling NtContinue directly from managed code like PowerShell, enabling stealthy endpoint evasion.

A signed Lenovo driver can be exploited in a Bring Your Own Vulnerable Driver (BYOVD) attack to terminate endpoint detection and response (EDR) processes. This technique allows attackers with elevated privileges to disable security monitoring, enabling further malicious activity without detection.

HDFC AMC disclosed a cybersecurity incident on May 16, 2026, after an anonymous source claimed access to portions of its IT infrastructure. The company activated containment and incident response protocols and engaged a specialist firm to assess impact. Initial assessment suggests no material operational impact, but the scope of access and data exposure remains unknown.

Researchers disclosed a new attack technique named GhostTree that can disrupt endpoint detection and response (EDR) tools on Windows systems, causing them to hang and leave files unscanned. This technique bypasses file scanning mechanisms, raising concerns for defenders about potential exploitation.

A new malware campaign targets macOS users with fake browser or system update prompts impersonating Apple, Google, and Microsoft. The malware steals passwords and other sensitive data from the endpoint. It highlights a social engineering technique that bypasses traditional security awareness by mimicking trusted update flows.

SentinelOne researchers discovered a macOS stealer named SHub Reaper that uses the applescript:// URL scheme to launch the Script Editor with a malicious payload, bypassing Terminal and Apple's TCC 26.4 mitigations. The malware spoofs Apple, Google, and Microsoft in a single attack chain to steal sensitive data.

At Pwn2Own Berlin 2024, security researchers disclosed 47 previously unknown zero-day vulnerabilities in various endpoint security and enterprise products. The event awarded $1.3 million in bounties, highlighting weaknesses in widely used security tools.

Fortinet confirmed unauthorized access to customer data stored in a third-party cloud-based file sharing environment. The breach did not involve encryption, ransomware, or compromise of Fortinet's corporate network, products, or source code. Exposed customer information could be used for phishing, social engineering, or impersonation attacks against affected enterprises.

A patched elevation-of-privilege vulnerability in the Windows Cloud Filter driver (cldflt.sys) has reappeared as a working SYSTEM-level exploit due to a regression. The flaw allows attackers to gain SYSTEM privileges, raising concerns about regression vulnerabilities in Windows.

MiniPlasma is a newly disclosed zero-day exploit that revives a 2020 vulnerability in the Windows cldflt.sys driver to achieve SYSTEM privilege escalation on fully patched Windows 11 systems as of May 2026. The technique bypasses existing patches, posing a risk to endpoint security by enabling attackers to gain elevated privileges.