Security news from the BlueTeamSec community for Monday the 9th of November 2026. A busy one today, with thirteen stories to work through.
Layer3 Intel have discovered residential proxy exit points originating from the White House network — 236 connections over a month from six distinct addresses. Either a device has been compromised or something's been inadvertently enrolled in a proxy network, likely via an installed SDK, which means external actors could be routing traffic straight through the perimeter. Follows up Brian Krebs' reporting from last week on the same incident.
Hunt Intelligence have exposed the infrastructure behind the South Korea bank hack that hit seven financial institutions between September and October. The attackers used an AI-driven autonomous penetration framework called ARTEX — powered by multiple LLM backends including Claude Code and DeepSeek — to compromise data on about sixty-seven thousand individuals and two thousand companies. Nineteen automated attack skills, rotating proxy infrastructure, and JNDI exploitation all orchestrated without direct human input.
And the US Department of Justice have indicted two Canadian nationals for a multi-year scheme targeting university supercomputers between 2016 and 2019. Ryan James Roach and Mathiew James Stubbings gained unauthorised access to high-performance computing environments and deployed malware for cryptocurrency mining, obfuscating their activities through compromised systems and VPS infrastructure.
Hayden McKenzie reports that North Korean IT worker syndicates are evolving their tactics — moving employer-issued devices from US laptop farms to Pakistan, whilst still using US-based VPN endpoints to mask their location. They're also increasingly using real local residents to pass identity verification during hiring, which is an adaptation to stepped-up law enforcement scrutiny. This follows multiple CISA alerts we've covered in recent weeks.
Acronis Threat Research have uncovered BOSSAI, a sophisticated attack framework run by a single Taiwan-based actor using a mesh of twelve AI agents to automate operations. The framework targeted hundreds of organisations across eleven countries, with nine agents running concurrently to handle infrastructure management, vulnerability scanning, and exploitation. Worth flagging if you're tracking how AI is being operationalised in offensive campaigns.
On the tooling front, researchers have released DefenderObserver, an open-source tool that monitors real-time telemetry from Microsoft Defender. It's designed to let red teamers and researchers observe how Defender reacts to specific inputs — essentially building a knowledge base for evasion technique development. One for those tracking offensive tradecraft or EDR visibility gaps.
A research tool called Keyeroo demonstrates an NTLM relay attack that bypasses LDAP signing on fully patched Domain Controllers. It exploits legacy NTLMv1 authentication to recover a victim's session key via Netlogon secure channel validation, enabling signed LDAP operations as the victim. Environments with compatibility level set to two or lower are vulnerable — though Windows 11 and Server 2025 mitigate this by dropping NTLMv1 support. Relevant if you're still running legacy authentication settings.
Check Point Research have identified a bypass for ChatGPT's Lockdown Mode that enables phishing, malware delivery, and unauthorised data exfiltration. It's another example of AI application security restrictions being circumvented — highlights emerging attack vectors specifically targeting AI platforms and their safety mechanisms.
Researchers have released HyperLoader, a proof-of-concept tool that executes shellcode within an isolated virtual CPU using the Windows Hypervisor Platform API. It evades userland EDR instrumentation by running malicious code in a hypervisor-isolated environment, bypassing traditional hooks placed in the host process. Maps host memory into the guest and proxies API calls to maintain functionality whilst hiding execution from security tools.
Florian Roth has written up what he describes as a tipping point — the moment when cyber attacks stop being limited by people. He's referring to AI-powered offensive frameworks like ARTEX and Cairn that can autonomously execute complete attack lifecycles without human oversight, operating around the clock and democratising advanced attack capabilities. The economic advantage shifts to attackers as inference costs decrease whilst defenders struggle with legacy software patch backlogs. Useful context piece if you're tracking the strategic implications of autonomous offensive tools.
Doctor Web report that threat actors are distributing multi-stage malware through poisoned GitHub repositories, specifically targeting information security professionals. The campaign uses malicious project files in security automation and testing tools to deploy an Electron-based loader that disables Defender and UAC, then delivers AsyncRAT, a clipper, and stealer payloads with advanced evasion including PPID spoofing and direct syscalls. We've seen similar developer-targeted campaigns in the past month — this one's notable for the specificity of the lure.
And CyberSecurity Malaysia have issued an advisory on a social engineering campaign targeting Malaysian banking customers. Fraudsters are impersonating Bank Negara Malaysia and police to trick victims into installing a malicious Android app called BNM Mobile Security. The banking trojan uses NFC relay attacks to exfiltrate contactless payment card data from major schemes and enables fraudulent transactions and ATM withdrawals.
Finally, Cyllex-Argus is a new AI-powered adversary intelligence framework for purple teams. It automates threat intelligence gathering and analysis using large language models to extract and consolidate intelligence from web sources, RSS feeds, and documents, mapping findings to MITRE ATT&CK. Notably, it employs a human-in-the-loop approach with quote-backed verification to prevent hallucinations and ensure traceability.
And that's your briefing. The articles are the work of their original authors — the analysis here was automated. Back tomorrow.