Good morning. This is your security briefing for Sunday, May 10, 2026, covering 19 articles. All attribution is by the article authors. All article analysis is automated. We begin with a critical alert from CISA.
CISA has added a critical SQL injection vulnerability in BerriAI's LiteLLM proxy server to its Known Exploited Vulnerabilities catalog. The flaw affects versions 1.81.16 through 1.83.6 and allows unauthenticated attackers to read and modify database contents including API keys and cloud credentials. Active exploitation has been observed in the wild within 36 hours of public disclosure. Organizations running LiteLLM must immediately upgrade to version 1.83.7 or later and review logs for indicators of compromise.
The United States Department of Justice announced that Deniss Zolotarjovs, a Latvian national and member of a prolific Russian ransomware organization, has been sentenced to 102 months in prison. Zolotarjovs specialized in extorting victims by analyzing stolen data and escalating pressure tactics, including exploiting children's health information from a pediatric healthcare company. The organization operated under multiple brands including Conti, Karakurt, Royal, and Akira, victimizing over 54 companies and compromising sensitive personal data of tens of thousands of individuals.
LAB52 has published analysis of a sophisticated cyber espionage campaign attributed to APT29, the Russian foreign intelligence service SVR. The campaign, dubbed EasterBunny, began in November 2025 and involved advanced malware deployment techniques in targeted attacks. The analysis provides insights into APT29's evolving tactics and procedures for espionage operations.
Cisco Talos has identified UAT-8302, a China-nexus advanced persistent threat group that has been targeting government entities in South America since late 2024 and southeastern Europe throughout 2025. The group appears focused on establishing long-term access for intelligence collection purposes.
The security blog r136a1 has published analysis noting a concerning trend in the cybersecurity community: the decline of public analyses of sophisticated, custom-built malware. Advanced malware research has increasingly moved behind paywalls and non-disclosure agreements, while public discourse is dominated by high-volume ransomware and infostealer reports. This shift obscures sophisticated state-sponsored and advanced persistent threat operations, creating intelligence blind spots for defenders who rely on public threat analysis.
SentinelOne has disclosed PCPJack, a sophisticated cloud worm and credential theft framework targeting exposed cloud infrastructure including Docker, Kubernetes, Redis, and MongoDB. The malware actively removes artifacts associated with the TeamPCP threat actor group before harvesting credentials, API keys, and sensitive data at scale, with worm-like propagation capabilities for rapid compromise across connected systems.
On May 8, 2026, Let's Encrypt temporarily shut down all certificate issuance due to a problem with the cross-signed certificate linking their Generation X root to their new Generation Y root. The organization resolved the issue by reverting all issuance back to the Generation X root certificate, affecting users of certain ACME certificate profiles.
DENIC, the registry for German domains, has published an analysis of a DNS outage that affected .de domains for approximately three hours on May 5, 2026. The issue stemmed from a flaw in DENIC's custom-developed DNSSEC signing component that generated three different key pairs with the same Key Tag, rendering only one-third of signature records valid and making domains unreachable. Despite monitoring tools detecting anomalies, alerts were not processed correctly, allowing the faulty zone to be published.
Microsoft has disclosed critical Remote Code Execution vulnerabilities in AI agent frameworks, particularly Microsoft Semantic Kernel. Two vulnerabilities allow attackers to exploit prompt injection to execute arbitrary code on host systems: one enables code execution through unsafe evaluation in Python lambda functions, and another allows arbitrary file writes to sensitive locations like Windows Startup folders. Developers and organizations using AI agent frameworks with plugins that interact with system resources are affected.
Security researcher p0dalirius has discovered two built-in command injection vulnerabilities in Windows context menus that enable arbitrary code execution. The vulnerabilities can be exploited through crafted file or folder names, shortcuts, or Explorer window elements that trigger context menu actions, with multiple attack scenarios demonstrating various injection vectors including payload embedding in filenames and user interface elements.
Progress has issued a critical security alert for MOVEit Automation addressing an authentication bypass vulnerability and a privilege escalation vulnerability. The flaws affect multiple versions of MOVEit Automation from 2024 and 2025 and could allow attackers to gain unauthorized access and administrative control. Patches are available but require full installer upgrades with system outages.
Researchers at 0xdeadbeefnetwork have disclosed Copy Fail 2, a local privilege escalation vulnerability affecting multiple Linux distributions including Ubuntu 24.04 LTS, Debian 13, Arch, and Fedora 43. The flaw exploits a mechanism in the network transformation layer to allow unprivileged users to modify the /etc/passwd file and gain root access without requiring sudo, enabling persistent root access through a modified user entry.
Red team researcher Mister-Joe has released PositiveIntent, a tool that enables obfuscation and in-memory execution of .NET assemblies while bypassing antimalware scan interface and event tracing security controls. The framework encrypts target assemblies, embeds them as resources, and reconstructs them in memory without writing to disk, designed specifically to evade modern endpoint detection and response systems.
SpecterOps research reveals that Dev Tunnels, a feature in Visual Studio Code and Cursor IDE, functions as an unintentional command and control framework with capabilities for remote code execution, persistence, lateral movement, and data exfiltration. The multi-layered protocol enables attackers to execute commands via remote procedure call methods, extract credentials from state database files, and leverage OAuth flows for initial access, affecting developers and organizations using these IDEs for remote development.
Security researchers at NasBench have discovered a technique to hijack DISM provider libraries using an undocumented sandbox flag. By replacing legitimate provider libraries with malicious ones in a user-specified directory, attackers can achieve arbitrary code execution within the DISM Host process context. This living-off-the-land technique leverages Windows built-in tooling for code execution without dropping traditional malware.
Researcher Azizcan DaΕtan has published SunnyDayBPF, a technique that uses extended Berkeley Packet Filter to perform post-syscall user-buffer deception, altering data after read-like system calls complete but before security agents can process it. This creates misleading telemetry that can fool downstream security systems including security information and event management platforms, endpoint detection and response, and audit backends, undermining the integrity of security monitoring by manipulating syscall data after collection.
Researcher afx underscore IDE is tracking a widespread ransomware campaign dubbed Sorry that is actively exploiting a critical authentication bypass vulnerability in cPanel and WHM to deploy ransomware against over 44,000 compromised IP addresses. Attackers force victims to tweet specific codes for confirmation and ransom notes are being publicly indexed by search engines, affecting various organizations including government entities.
A researcher at eversinc33 has documented building a custom devirtualizer using the LLVM C++ API to reverse-engineer virtualized crackme software. The tool lifts virtualized bytecode into LLVM intermediate representation, applies optimization passes to strip away obfuscation layers like those used by Themida or VMProtect, and exposes original program logic, demonstrating how to defeat virtualization-based code protection through compiler optimization.
BlueVoyant has identified a sophisticated, well-funded threat group distributing the Lorem Ipsum multi-stage loader and backdoor through SEO-poisoned trojanized Microsoft Teams installers since February 2026. The malware employs advanced evasion techniques including validly code-signed MSI installers, library sideloading, AES encryption with multiple obfuscation layers, and command and control traffic disguised as image files. The threat actors demonstrate rapid development velocity suggesting potential AI-assisted tooling.
Security researcher gsmll has released HyperVenom, a framework that achieves Ring negative-1 execution by hijacking Microsoft Hyper-V during boot. Using a UEFI bootloader to inject payloads into the hypervisor address space, the framework intercepts VM-exits to perform deep memory introspection and manipulation from usermode, bypassing Ring 0 security measures. It targets Windows systems with Hyper-V and Virtualization-Based Security enabled, designed to evade kernel-level anti-cheat and telemetry systems.
That concludes today's briefing.