Good morning. This is your security briefing for Thursday, April 09, 2026, covering 15 articles. All attribution is by the article authors. All article analysis is automated.
The Justice Department and FBI disrupted a DNS hijacking network operated by Russia's GRU Military Unit 26165, also known as APT 28. According to the Department of Justice, the operation exploited vulnerabilities in TP-Link routers worldwide to conduct adversary-in-the-middle attacks, harvesting credentials and authentication tokens from military, government, and critical infrastructure targets. The UK National Cyber Security Centre issued an advisory about the campaign, and Infrawatch has released a checker tool that allows organizations to determine if their IP address or network may have been compromised by analyzing public addresses with hourly refreshed data.
Cisco Talos identified LucidRook, a new Lua-based malware stager used by threat actor UAT-10362 in targeted spear-phishing campaigns against Taiwanese NGOs and universities. The malware embeds a Lua interpreter and Rust libraries, downloads staged payloads via compromised FTP servers, and features layered anti-analysis capabilities, while a companion tool called LucidKnight exfiltrates system information via Gmail for target profiling.
Trellix ARC reported on Masjesu, a commercially operated IoT botnet active since early 2023 that functions as a DDoS-for-hire service advertised through Telegram. The botnet targets diverse IoT devices across multiple architectures, employs encryption for obfuscation, and uses advanced persistence mechanisms including process spoofing and cron jobs while deliberately avoiding sensitive IP ranges to evade detection, with demonstrated attack capabilities reaching approximately 290 gigabits per second.
Tracebit described using canary credentials as a detection mechanism for CI/CD supply chain attacks. Canary credentials are intentionally placed fake, low-privilege credentials that trigger alerts when accessed, providing early warning of pipeline compromise and allowing organizations to identify unauthorized access before significant damage occurs.
A formal verification study by Cobalt AI researchers analyzed 3,500 code artifacts from seven leading AI coding assistants including GPT-4o, Claude, Gemini, Mistral, and Llama models, finding that nearly 56 percent contained at least one security vulnerability. Using formal proof methods, researchers identified over one thousand exploitable vulnerabilities across critical categories including memory allocation, integer arithmetic, authentication, and cryptography, with all tested AI models demonstrating high vulnerability rates and none achieving better than a D security grade.
A security researcher disclosed a null-signature authentication bypass vulnerability in NSA's Ghidra Server PKI authentication module, affecting all versions through 12.0.4 since the initial open-source release in March 2019. The flaw incorrectly skips client signature verification when signature bytes are null, allowing users with valid CA-signed certificates to impersonate other users and gain unauthorized access to collaborative reverse engineering projects.
Cisco Talos warned that cybercriminals are weaponizing SaaS notification pipelines from platforms like GitHub and Jira in a tactic called Platform-as-a-Proxy. Attackers exploit legitimate notification features by embedding malicious links in commit messages or service desk notifications, allowing them to bypass traditional email security filters since these communications pass standard authentication checks, enabling credential harvesting and social engineering attacks.
Binary Defense researchers highlighted that attackers are increasingly weaponizing Scalable Vector Graphics files in phishing campaigns by embedding malicious JavaScript within these XML-based files. The technique exploits the perceived harmlessness of image files to bypass traditional security measures and execute code that harvests credentials or delivers malware, with researchers recommending defenders shift from signature-based detection to behavioral hunting.
A new tool called mssqlbof was publicly released, providing a Beacon Object File suite designed for red team operations against Microsoft SQL Server environments. It implements the database protocol directly in C to enable stealthy SQL Server interactions without loading detectable libraries, PowerShell, or .NET, and supports advanced post-exploitation capabilities including remote command execution, login impersonation, privilege escalation enumeration, credential dumping, and pass-the-hash authentication.
Researchers discovered Environment-injected Trajectory-based Agent Memory Poisoning, a novel attack technique targeting memory-augmented web agents and AI browsers. Attackers embed malicious instructions in web content that poison agent memory during one task, then activate during semantically related tasks on different sites, enabling cross-session and cross-site compromise while bypassing permission-based defenses.
Push Security reported that device code phishing attacks exploiting OAuth 2.0 Device Authorization Grant have increased over 37 times in 2026, targeting users of Microsoft, Google, Salesforce, GitHub, and AWS. Attackers use legitimate device code login pages to trick victims into authorizing malicious applications, bypassing multi-factor authentication and passkeys, with attack kits like EvilTokens being used to automate the process.
Sekoia detailed EvilTokens, a new AI-augmented Phishing-as-a-Service platform that automates Business Email Compromise fraud. The platform uses AI to analyze harvested emails, identify financial opportunities, and craft convincing phishing lures, while employing device code phishing kits to steal Microsoft account tokens for data exfiltration and persistence, with the entire operation controlled through Telegram bots and post-compromise activities occurring within minutes.
A publicly released tool called AgentMemshell enables injection of memory-resident shells into Java processes across all JDK versions, specifically targeting Tomcat middleware. The tool is designed for scenarios with no command execution feedback, allowing attackers to establish persistent, difficult-to-detect access within Java applications by residing entirely in memory.
Researchers from The Hong Kong Polytechnic University and partner institutions introduced Argus, a novel large language model-centered multi-agent framework for full-chain security vulnerability detection that addresses limitations in traditional static analysis tools. The framework reorchestrates static analysis workflows using language models as primary agents to detect sophisticated vulnerabilities with improved accuracy and reduced false positives, and has already identified several critical zero-day vulnerabilities that received CVE assignments.
That concludes today's briefing.