Threat Research | Weekly Recap [03 May 2026]

AI, SaaS, and Collaboration Threats

• AI assistants, agents, and collaboration tools remain high-risk as prompt injection, OAuth abuse, and multi-channel incidents rise; AI and Human Risk Landscape
• Trojanized models, datasets, and agent skills on AI platforms enabled hidden payloads, indirect prompt injection, and covert C2; AI supply chain attacks on Hugging Face and OpenClaw
• High-risk “AI productivity” browser extensions abused browser privileges for RAT, infostealer, proxy hijacking, and AitB activity; AI browser extensions reading emails first
• PromptMink used malicious npm dependencies and AI-generated commits to exfiltrate secrets and crypto wallet access; Claude adds malware to crypto agent
• AI-native analytics and autonomous agents are being used to correlate telemetry and accelerate detection-to-containment; Cyble Blaze AI Threat Intelligence Automation

Phishing, BEC, and IdP/SaaS Intrusions

• QR-code and CAPTCHA-gated phishing, plus persistent BEC, dominated email threats while Tycoon2FA was briefly disrupted; Q1 2026 Email Threat Landscape
• Vishing-led AiTM pages and adversary-registered MFA devices enabled rapid SaaS takeovers and data theft; CORDIAL SPIDER and SNARKY SPIDER
• A fake Meta login flow used Google Forms and a vercel.app page to harvest credentials and 2FA codes in real time; Meta 2FA trap
• Kuse.ai sharing was abused to host a phishing document that redirected victims to a fake Microsoft login; Kuse web app phishing abuse

Supply Chain and Developer Ecosystem Abuse

• Malicious Ruby gems and Go modules impersonated dev tools to steal secrets, poison CI, and persist in build environments; BufferZoneCorp package campaign
• A brand-squatted TanStack npm package exfiltrated environment files via postinstall scripts; TanStack npm brand-squat
• CI/CD workflow tampering detection was highlighted with LLM-augmented rules across GitHub Actions, GitLab CI, and Azure DevOps; CI/CD pipeline abuse detector
• Malicious npm dependencies and AI-generated commits were used to plant backdoors and steal crypto assets; PromptMink supply-chain malware

Ransomware, Backdoors, and Malware Campaigns

• Vect RaaS expanded across Windows, Linux, and ESXi with ChaCha20 encryption and broad lateral movement; Inside Vect RaaS
• VECT 2.0’s flawed encryption can permanently destroy larger files, effectively behaving like a wiper; VECT ransomware by design, wiper by accident
• Silver Fox used tax-themed phishing to deploy ValleyRAT and the new ABCDoor backdoor with persistence tricks; Silver Fox and ABCDoor
• A trojanized CHM sample delivered a Python-based Telegram RAT with Winlogon hijack and scheduled-task persistence; Rebex-based Telegram RAT
• ClickFix/BackgroundFix staged CastleLoader to drop NetSupport RAT and CastleStealer for credential and wallet theft; ClickFix leaves the malware

Cloud, Kubernetes, and SaaS Security

• Kubernetes attack detection coverage was expanded with Wazuh guidance and telemetry for cluster-focused threats; Detecting Kubernetes attacks with Wazuh
• Chinese threat actors increasingly route activity through compromised SOHO/IoT edge nodes and shared malware ecosystems; Chinese threat actors overview
• UAT-8099 leveraged vulnerable IIS servers, web shells, PowerShell, and BadIIS variants against regional targets; DNS facts on UAT-8099
• LiteLLM was rapidly targeted with pre-auth SQL injection to enumerate API keys, provider credentials, and environment variables; CVE-2026-42208 LiteLLM SQLi

Vulnerability Research and DFIR

• CVE-2026-31431 lets local users gain root via a Linux kernel crypto interface flaw; Copy Fail kernel flaw
• GitHub Enterprise Server and GitHub cloud fixed an RCE issue tied to malformed git push options; CVE-2026-3854 GitHub RCE
• Osquery and Elastic Security showed how to trace phishing-delivered Mimikatz execution without full disk imaging; DFIR with Osquery in Elastic Security

Threat Intel, Infrastructure, and Investigation Tooling

• Behaviorally fingerprinted infrastructure graphs and pre-correlated signals help SOCs cluster adversary infrastructure faster; Cluster adversary infrastructure faster
• Validin launched a new Advanced Search experience with VQL, pivots, and same-observation matching for investigations; Validin Advanced Search

Threat Research | Weekly Recap – hendryadrian.com