Endpoint Security Signal — Week of August 31, 2026

2026-09-06

incident

Kimsuky Evolves Operation GitPower with AI-Generated Decoys and Expanded C2 InfrastructureOperation GitPower

North Korean state-sponsored group Kimsuky is upgrading its Operation GitPower campaign by using AI agents to mass-produce targeted phishing decoys and diversifying command-and-control infrastructure across GitHub and Pastebin. The campaign targets individuals and organizations, likely for espionage, and the use of AI-generated content makes phishing more convincing and harder to detect.

↗ brinztech.com (2026-09-06) ↗ koreatimes.co.kr (2026-09-07) ↗ gbhackers.com (2026-09-07)

2026-09-06

research poc

Researchers demonstrate abuse of Windows Security Center APIs to disable security toolsSecurity Center API abuse

Researchers demonstrated a technique where threat actors abuse legitimate Windows Security Center APIs to programmatically register a fake antivirus, causing Windows to disable real security tools. This defensive evasion tactic shifts from loud process termination to sophisticated API manipulation, posing a significant risk to endpoint protection.

↗ aisecuritywatch.com (2026-09-06)

2026-09-06

vuln disclosure

JetBrains TeamCity Vulnerability CVE-2026-63077 Exploited to Exfiltrate AWS Credentials and Source CodeCVE-2026-63077

Attackers exploited an unpatched vulnerability in JetBrains TeamCity (CVE-2026-63077) to exfiltrate AWS credentials and source code from Cadence. The breach highlights the risk of unpatched CI/CD servers and the potential for supply chain compromise.

JetBrains ↗ rescana.com (2026-09-06)

2026-09-05

research poc

AI Agents Breach Enterprise Network in Under 10 Hours, Steal Root CredentialsAI-Agent-Assisted Breach

A red team exercise using frontier AI models breached an enterprise network and obtained root credentials in under 10 hours, a task that typically takes human red teams about two weeks. The AI agents autonomously performed reconnaissance, exploitation, and privilege escalation, demonstrating a significant acceleration in attack speed.

↗ cybersecuritynews.com (2026-09-05) ↗ news-pravda.com (2026-09-06) ↗ siliconangle.com (2026-09-08) ↗ esecurityplanet.com (2026-09-08) ↗ privacyneedle.com (2026-09-10) ↗ hivesecurity.gitlab.io (2026-09-10) ↗ privacyneedle.com (2026-09-10)

2026-09-04

research poc

Toy Ghouls Uses Messaging Services to Control Windows BackdoorsToy Ghouls

A threat actor tracked as Toy Ghouls is deploying custom Windows backdoors that use popular messaging services for command and control, allowing them to maintain control over compromised systems while evading detection. The backdoors are designed to blend in with legitimate traffic, making them harder to detect by security tools.

↗ cybersecuritynews.com (2026-09-04) ↗ gbhackers.com (2026-09-04) ↗ teamwin.in (2026-09-05)

2026-09-04

research poc

NodeStealer Spyware Adds Keylogging, Screenshot Capture and Facebook Data TheftNodeStealer

NodeStealer, a Python-based infostealer, has been upgraded with keylogging, screenshot capture, and expanded Facebook data theft capabilities. The malware now functions as a broader spyware platform, increasing its threat to endpoint security.

↗ gbhackers.com (2026-09-04)

2026-09-04

vuln disclosure

Medium Severity Vulnerability CVE-2026-85149 DisclosedCVE-2026-85149

A medium severity vulnerability tracked as CVE-2026-85149 has been disclosed. The article does not specify the affected vendor, product, or technical details. It advises applying vendor patches and checking official advisories.

↗ thehackerwire.com (2026-09-04) ↗ thehackerwire.com (2026-09-07) ↗ thehackerwire.com (2026-09-10) ↗ thehackerwire.com (2026-09-11) ↗ thehackerwire.com (2026-09-13)

2026-09-04

research poc

GPT-6 Astra Scores 100% on ExploitBench; OpenAI Restricts Exploit RequestsGPT-6 Astra ExploitBench

OpenAI's GPT-6 Astra model achieved a perfect score on ExploitBench, a benchmark for automated exploit generation, demonstrating the ability to develop functional exploits, including two zero-days. In response, OpenAI has restricted the model to code review and patching tasks, blocking requests for proof-of-concept exploits.

↗ thehackernews.com (2026-09-04) ↗ etnownews.com (2026-09-04) ↗ penligent.ai (2026-09-05)

2026-09-04

vuln disclosure

Nightmare Eclipse discloses privilege escalation vulnerabilities in Kaspersky, Avast, and NvidiaNightmare Eclipse privilege escalation vulnerabilities

Researcher Nightmare Eclipse published privilege-escalation vulnerabilities affecting Kaspersky and Avast security products, plus a separate Nvidia issue. Independent researcher Kevin Beaumont reportedly validated the CrowdStrike, Kaspersky, and Avast proofs-of-concept, increasing confidence in the findings.

Kaspersky · Avast · Nvidia ↗ bragg.substack.com (2026-09-04) ↗ coesecurity.com (2026-09-07) ↗ securityweek.com (2026-09-07) ↗ igorslab.de (2026-09-08) ↗ hendryadrian.com (2026-09-08) ↗ coesecurity.com (2026-09-07)

2026-09-03

vendor announcement

Sophos Integrates OpenAI GPT Models into Managed Risk for Exploit Path VerificationSophos-EPV-GPT

Sophos announced Exploit Path Verification (EPV), a new capability in its Managed Risk offering that uses OpenAI GPT models to validate exploit paths. This helps defenders prioritize and remediate risks by confirming whether identified attack paths are actually exploitable.

Sophos ↗ markets.businessinsider.com (2026-09-03) ↗ manilatimes.net (2026-09-03)

2026-09-03

incident

Threat actors use LLMs and RATs in campaign against Mexican and Brazilian organizationsLLM-assisted RAT campaign targeting Mexico and Brazil

A cyber campaign targeting organizations in Mexico and Brazil leverages Large Language Models (LLMs), Remote Access Trojans (RATs), and SOCKS5 proxies to establish evasive command-and-control and persistent access. The use of LLMs likely aids in crafting convincing phishing lures or automating attack components. The campaign demonstrates evolving tactics to bypass endpoint defenses.

↗ brinztech.com (2026-09-03) ↗ teamwin.in (2026-09-05) ↗ securityonline.info (2026-09-09) ↗ gbhackers.com (2026-09-10)

2026-09-03

research poc

Attackers Abuse Signed Node.js Binaries and Ethereum Smart Contracts for Evasive C2EtherHiding

A malware campaign abuses digitally signed Node.js binaries and Ethereum smart contracts to establish persistent, evasive command-and-control (C2) infrastructure. The technique, known as EtherHiding, leverages blockchain immutability to hide C2 addresses, making takedown difficult. This poses a significant challenge for endpoint security products that rely on detecting malicious binaries or network indicators.

↗ brinztech.com (2026-09-03)

2026-09-03

research poc

EtherHiding Upgrades Amatera Stealer with WebRTC-based C2Amatera Stealer WebRTC C2

The threat actor EtherHiding has upgraded the Amatera infostealer to use a custom WebRTC channel for command-and-control, bypassing traditional signaling servers. This makes network detection harder because C2 traffic blends with legitimate WebRTC usage.

↗ brinztech.com (2026-09-03)

2026-09-03

incident

China-linked hackers backdoored executives' laptops via USB boot attacksUSB boot attack

Chinese state-linked hackers compromised executive laptops at a Hainan conference using USB boot attacks, bypassing traditional security measures. The attackers exploited a fix that companies had but weren't using, indicating a failure in patch deployment. This incident highlights the risk of physical access attacks against endpoints.

↗ venturebeat.com (2026-09-03)

2026-09-03

vendor announcement

CrowdStrike Announces Agentic Identity Provider at Fal.ConCrowdStrike Agentic Identity Provider

CrowdStrike announced its Agentic Identity Provider at Fal.Con, aiming to close a gap in its Continuous Identity offering. The new product integrates identity protection with the Falcon platform to enhance security for enterprise customers.

CrowdStrike ↗ siliconangle.com (2026-09-03)

2026-09-03

vuln disclosure

Chaotic Eclipse Claims Avast Antivirus 0-Day Vulnerability with PoC ReleaseAvast Antivirus 0-Day

A threat actor known as Chaotic Eclipse has publicly claimed a zero-day vulnerability in Avast Antivirus and released a proof-of-concept exploit. The claim has not been independently verified by Avast or assigned a CVE.

Avast ↗ cybersecuritynews.com (2026-09-03) ↗ cybersecuritynews.com (2026-09-03)

2026-09-03

vendor announcement

Microsoft to Automatically Enable Memory Integrity on Windows DevicesMemory Integrity Auto-Enablement

Microsoft announced that starting October 2026, it will automatically enable Memory Integrity (HVCI) on eligible Windows devices through quality updates. This change aims to block kernel-level attacks by enforcing code integrity in the kernel. The move is part of broader security enhancements to Windows.

↗ gbhackers.com (2026-09-03)

2026-09-03

vuln disclosure

VMware Workstation and Fusion Vulnerabilities Allow Host Code ExecutionCVE-2025-22226

VMware disclosed two vulnerabilities in Workstation and Fusion that could allow an attacker with local administrative privileges inside a virtual machine to execute code on the host. The flaws are information disclosure and out-of-bounds read issues in the HGFS and shader components. Patches are available.

VMware ↗ cybersecuritynews.com (2026-09-03) ↗ securityaffairs.com (2026-09-05)

2026-09-03

research poc

Avast Antivirus Zero-Day PoC Enables SAM Database Dump and SYSTEM ShellMSNightmare

A proof-of-concept exploit named MSNightmare targets a zero-day vulnerability in Avast Antivirus, allowing attackers to dump the SAM database and obtain a SYSTEM shell. The PoC has been tested against patched versions of Avast and Windows 11 25H2, and includes source code and compiled binaries. No CVE identifier has been assigned yet.

Avast ↗ gbhackers.com (2026-09-03)

2026-09-03

incident

BREEZE COMET Hackers Use AI-Assisted Malware to Target Brazil BanksBREEZE COMET

A threat group named BREEZE COMET is using AI-assisted malware to target Brazilian banks for fraudulent transfers. The malware is designed to evade detection and manipulate banking transactions, posing a significant threat to financial institutions.

↗ teamwin.in (2026-09-03)

2026-09-03

vuln disclosure

CrowdStrike Falcon Sensor Local Privilege Escalation Zero-Day (FalconFlank)FalconFlank

A proof-of-concept exploit named FalconFlank has been publicly released, claiming to achieve local privilege escalation on systems running CrowdStrike Falcon Sensor. The vulnerability is a zero-day affecting the Falcon sensor, potentially allowing attackers to elevate privileges on compromised endpoints. This disclosure highlights a critical security gap in a widely deployed endpoint detection and response (EDR) product.

2026-09-03

vuln disclosure

SonicWall discloses actively exploited SMA1000 vulnerabilitiesCVE-2025-23006

SonicWall disclosed two vulnerabilities in its Secure Mobile Access (SMA) 1000 series, including a critical pre-authentication deserialization flaw (CVE-2025-23006) that allows unauthenticated remote code execution. The vulnerabilities are under active exploitation, and SonicWall has released patches but no workarounds.

SonicWall ↗ csoonline.com (2026-09-03)

2026-09-03

vendor announcement

CrowdStrike Adds Real-Time Supply Chain Attack ProtectionCrowdStrike Real-Time Supply Chain Attack Protection

CrowdStrike announced a new capability that provides real-time protection against software supply chain attacks by blocking malicious packages at endpoints before embedded scripts can execute. This enhancement aims to prevent attackers from compromising development pipelines and distributing malicious code to endpoints.

↗ channelinsider.com (2026-09-03) ↗ cioinfluence.com (2026-09-03) ↗ cyberdaily.au (2026-09-03) ↗ msspalert.com (2026-09-04)

2026-09-02

research poc

Constructing a CrowdStrike EDR Evasion Debugging PipelineEDR Evasion Debugging Pipeline

A security researcher details a reproducible pipeline for testing EDR evasion techniques against CrowdStrike Falcon. The pipeline uses a custom tool called Detonator to automate payload execution and capture detection telemetry, enabling iterative refinement of evasion methods. This research aims to improve defensive understanding by exposing how evasion techniques interact with EDR internals.

CrowdStrike ↗ dev.to (2026-09-02)

2026-09-02

research poc

AI Agents Execute Full Ransomware Attack in Under 10 HoursAI-Agent Ransomware Attack

Unit 42 researchers documented a ransomware attack where AI agents autonomously performed every step, from initial access to encryption, in under 10 hours. The attack would normally take human operators about two weeks. The AI agents also left behind an 80-page security audit of the victim's network.

↗ theregister.com (2026-09-02) ↗ zdnet.com (2026-09-02) ↗ techtimes.com (2026-09-03) ↗ scworld.com (2026-09-03) ↗ csoonline.com (2026-09-03)

2026-09-02

research poc

RecoverIt: Using Windows Service Failure Recovery for Defense EvasionRecoverIt

Security researchers have documented a technique called RecoverIt that abuses Windows Service Failure Recovery to achieve persistence and evade detection. The method modifies an existing service's failure recovery settings to execute a payload when the service fails, without creating a new service. This technique can bypass endpoint security controls that monitor for new service creation or suspicious registry modifications.

↗ hackers-arise.com (2026-09-02)

2026-09-02

research poc

MoiClient evasive backdoor uses DLL side-loading, process injection, UAC bypass, and BYOVDMoiClient

MoiClient is an evasive backdoor distributed via invoice-lure VHDX files. It employs DLL side-loading, process injection, RPC-based UAC bypass, and BYOVD to evade defenses and maintain persistence. The malware reappears via Task Scheduler and ultimately delivers MoiXD Stealer.

↗ hendryadrian.com (2026-09-02)

2026-09-02

vendor announcement

CrowdStrike announces integrations with Commvault, Rubrik, and VAST Data for cyber-recoveryCrowdStrike-Commvault-Rubrik-VAST-Data-integration

At Fal.Con 2026, CrowdStrike announced new integrations between its security agents and cyber-recovery platforms from Commvault, Rubrik, and VAST Data. These integrations allow CrowdStrike's detection capabilities to trigger recovery workflows in the partner platforms, aiming to accelerate response to ransomware and other attacks.

CrowdStrike · Commvault · Rubrik · VAST Data ↗ blocksandfiles.com (2026-09-02)

2026-09-02

research poc

Singularity Rootkit Bypasses Elastic Defend eBPF Module Load DetectionSingularity Rootkit eBPF Module Load Detection Bypass

A security researcher disclosed a technique used by the Singularity Linux rootkit to evade Elastic Defend by suppressing module-load telemetry. The method prevents the endpoint security agent from detecting the rootkit's kernel module loading, allowing it to operate undetected across multiple layers.

Elastic ↗ gbhackers.com (2026-09-02)

2026-09-02

research poc

TukTuk malware steals credentials and disables security toolsTukTuk

Researchers discovered TukTuk, a malware framework used by ransomware operators to steal credentials and disable endpoint security tools. The framework was recovered from a server containing malicious DLL sideloading sets and EDR-disabling components, providing insight into ransomware infrastructure.

↗ cryptika.com (2026-09-02) ↗ gbhackers.com (2026-09-02) ↗ teamwin.in (2026-09-03)

2026-09-02

vendor announcement

OpenAI developing AI model capable of finding unknown vulnerabilities and developing exploitsOpenAI advanced cyber model

OpenAI is developing an advanced AI model that can reportedly find previously unknown vulnerabilities and develop exploits across well-protected systems without human guidance. The company delayed parts of development to strengthen safeguards and plans to restrict advanced cyber access. This development has implications for endpoint security as AI-driven exploit generation could challenge existing defenses.

↗ anothercodingblog.com (2026-09-02) ↗ dev.to (2026-09-12)

2026-09-02

research poc

AI-Assisted Cyber Attack Investigation by Unit 42AI-Assisted Cyber Attack

Unit 42 investigated an AI-assisted cyber attack where autonomous AI agents breached an enterprise network in hours. The attack demonstrates the use of AI to accelerate and automate intrusion. This highlights emerging threats from agentic AI in offensive operations.

↗ unit42.paloaltonetworks.com (2026-09-02) ↗ europesays.com (2026-09-02) ↗ gbhackers.com (2026-09-05) ↗ techzine.eu (2026-09-08)

2026-09-02

incident

Attackers Conceal Reverse Shell Traffic via Signed Apps and AWS API GatewaySigned App + AWS API Gateway C2

Threat actors are conducting a fake IT-support campaign to gain remote access, deploy a malicious MSI package, and hide hands-on-keyboard activity. They use signed applications and AWS API Gateway to blend command-and-control traffic with legitimate cloud services, evading network detection.

↗ gbhackers.com (2026-09-02)

2026-09-02

research poc

Ransomware Group Builds Custom C2 Framework to Steal Passwords and Disable Security ToolsThe Gentlemen C2 Framework

A server linked to the ransomware group The Gentlemen hosted a previously undocumented command-and-control (C2) framework. The framework is designed to steal passwords and kill security tools, indicating a custom-built tool for post-exploitation and defense evasion.

↗ cyberpress.org (2026-09-02)

2026-09-01

incident

Law enforcement and CrowdStrike disrupt Sality botnet via sinkholingSality botnet disruption

International law enforcement, CrowdStrike, and Shadowserver Foundation disrupted the Sality peer-to-peer botnet, which had infected over 15,000 machines. The operation poisoned the botnet's peer list and diverted traffic to sinkholes, preventing further command-and-control. Sality is a 23-year-old malware family known for delivering additional payloads.

↗ theregister.com (2026-09-01) ↗ thenextweb.com (2026-09-02)

2026-09-01

vendor announcement

OpenAI announces Astra, an LLM for finding and exploiting security vulnerabilitiesOpenAI Astra

OpenAI announced Astra, its first LLM designed to find and exploit security vulnerabilities. The model is intended for defensive security research and can identify and exploit flaws in software. This marks a significant step in applying LLMs to offensive security tasks.

↗ ecosistemastartup.com (2026-09-01) ↗ decrypt.co (2026-09-02) ↗ moneycheck.com (2026-09-02) ↗ firstpost.com (2026-09-02) ↗ techtimes.com (2026-09-02)

2026-09-01

vendor announcement

CrowdStrike Announces SafeMind Agentic AI Security System at Fal.ConCrowdStrike SafeMind

CrowdStrike announced SafeMind, its first complete agentic system for cybersecurity, at Fal.Con. SafeMind combines a model, a specialized harness, and a digital twin of an enterprise environment to simulate attacks and defenses. This enables AI-driven red teaming and blue teaming to identify vulnerabilities and improve security posture.

CrowdStrike ↗ investing.com (2026-09-01) ↗ csoonline.com (2026-09-02) ↗ techzine.eu (2026-09-02) ↗ esecurityplanet.com (2026-09-02) ↗ scworld.com (2026-09-02) ↗ securitybrief.news (2026-09-04) ↗ ciso.economictimes.indiatimes.com (2026-09-05) ↗ simplywall.st (2026-09-05) ↗ hipther.com (2026-09-07)

2026-09-01

research poc

Gentlemen Ransomware Group Uses DLL Side-Loading and GentleKiller to Evade EDRGentleKiller

The Gentlemen ransomware group uses DLL side-loading via legitimate binaries like Greenshot and SyncTrayzor to execute malicious payloads in trusted process contexts. They also deploy a custom tool called GentleKiller to neutralize EDR solutions, disabling them before encryption. This research exposes their C2 framework and evasion tactics.

↗ www.brinztech.com (2026-09-01) ↗ decryptiondigest.com (2026-09-03) ↗ cybersecuritynews.com (2026-09-03) ↗ cyberpress.org (2026-09-03) ↗ securityonline.info (2026-09-04)

2026-09-01

research poc

AzaleaControl RAT Emerges with Rootkit and EDR EvasionAzaleaControl

A new commercial remote access trojan (RAT) named AzaleaControl is being advertised on underground forums. It offers modular post-exploitation capabilities including a Ring3 rootkit, indirect syscalls, Hidden VNC, and Active Directory enumeration. The tool is designed to evade endpoint detection and response (EDR) systems.

↗ brinztech.com (2026-09-01)

2026-09-01

incident

Anthropic Force-Logs Out Users After Infostealer Malware Hijacks Claude AI SessionsInfostealer hijacking of Claude AI sessions

Anthropic force-logged out thousands of Claude users after infostealer malware compromised their AI sessions. The malware stole authentication tokens, allowing attackers to hijack active Claude sessions. This incident highlights the growing threat of infostealers targeting AI service credentials.

Anthropic ↗ blog.lufsec.com (2026-09-01) ↗ www.cyberkendra.com (2026-09-11)

2026-09-01

incident

Threat Actors Abuse ChatGPT Shared Links to Distribute NetSupport RATChatGPT Shared Links Abuse for NetSupport RAT

Threat actors are abusing legitimate ChatGPT shared-conversation URLs to host social-engineering lures that lead to a ClickFix-style infection chain, ultimately deploying the NetSupport RAT. This technique leverages the trust and reputation of ChatGPT domains to bypass security controls and trick users into executing malicious commands.

↗ gbhackers.com (2026-09-01)

2026-09-01

incident

BraZetsu AI Malware Ranks and Sells Access to Compromised Corporate NetworksBraZetsu

BraZetsu malware targets Brazil and Latin America, mapping corporate systems and using AI to rank compromised networks for sale to other criminals. This incident highlights the use of AI in malware to automate network reconnaissance and monetize access.

↗ cybersecuritynews.com (2026-09-01) ↗ securityonline.info (2026-09-04)

2026-09-01

vendor announcement

CrowdStrike Launches Falcon Guardian to Block Unauthorized AI AgentsFalcon Guardian

CrowdStrike announced Falcon Guardian, a new capability that inventories and blocks unauthorized AI agents on Windows and macOS endpoints. It traces agent activity including prompts, identities, and tool calls to downstream actions, aiming to prevent unapproved AI agents from operating.

↗ stocktitan.net (2026-09-01) ↗ siliconangle.com (2026-09-01) ↗ cloudnews.tech (2026-09-01) ↗ techzine.eu (2026-09-01) ↗ securityboulevard.com (2026-09-01) ↗ cyberdaily.au (2026-09-02) ↗ seekingalpha.com (2026-09-02) ↗ securitybrief.com.au (2026-09-03) ↗ chatgptaihub.com (2026-09-05) ↗ questreviewcenter.com (2026-09-06) ↗ ca.finance.yahoo.com (2026-09-06) ↗ cryptobriefing.com (2026-09-08)

2026-09-01

research poc

AI Agent Orchestrates Entire Ransomware Attack in JADEPUFFER CampaignJADEPUFFER

JADEPUFFER is described as the first ransomware campaign run entirely by an AI agent, autonomously breaching and destroying a database. This signals a new threat era where AI agents can orchestrate full attack chains without human intervention.

↗ telecomreviewasia.com (2026-09-01) ↗ hivesecurity.gitlab.io (2026-09-10)

2026-09-01

incident

Fake Cloudflare CAPTCHA tricks victims into opening a tunnel for attackersTerminalFix

Attackers are using a fake Cloudflare CAPTCHA to trick victims into running malicious PowerShell commands, launching a multi-stage attack that ends with a reverse tunnel into the compromised network. The campaign, dubbed TerminalFix, uses social engineering to bypass endpoint security and establish persistent access.

↗ csoonline.com (2026-09-01) ↗ hothardware.com (2026-09-01)

2026-08-31

research poc

Attack hides malware in PNGs and drops custom reverse tunnelPNG-based malware delivery with custom reverse tunnel

A new ClickFix campaign uses PNG images to hide malicious payloads and deploys a custom reverse tunnel for command and control. The multi-stage attack chain evades detection by embedding malware in image files and using legitimate tools for execution.

↗ theregister.com (2026-08-31)

2026-08-31

incident

Aurora Ransomware Operators Use Cursor AI in Attacks on 10 OrganizationsAurora Ransomware AI-Assisted Attack

The Aurora ransomware group leveraged the Cursor AI coding assistant, powered by Claude Sonnet, to direct post-compromise reconnaissance, lateral movement, and exploitation across at least 10 corporate networks. This marks a notable use of AI assistance in ransomware operations, potentially lowering the skill barrier for attackers.

↗ brinztech.com (2026-08-31)

2026-08-31

research poc

HardBreacher PoC Exploits Kaspersky Endpoint Security Zero-Day for Windows 11 Privilege EscalationHardBreacher

A proof-of-concept exploit called HardBreacher allegedly exploits an unpatched local privilege escalation vulnerability in Kaspersky Antivirus for Endpoint. The flaw allows a local user to control a privileged component, potentially leading to full system compromise. The vulnerability affects Kaspersky Endpoint Security on Windows 11.

Kaspersky ↗ gbhackers.com (2026-08-31) ↗ cybersecuritynews.com (2026-08-31) ↗ blacktree.nl (2026-08-31) ↗ cybersecuritynews.com (2026-08-31) ↗ news-pravda.com (2026-09-01) ↗ teamwin.in (2026-09-01) ↗ webboard-nsoc.ncsa.or.th (2026-09-01) ↗ www.show.it (2026-09-01) ↗ cybersecuritynews.com (2026-09-07)

2026-08-31

research poc

Researchers Demonstrate AI Agents Executing Malicious Commands from llms.txt Filesllms.txt poisoning

Researchers scanned 6,214 corporate domains and found 120 llms.txt files containing 227 commands pointing to unregistered packages. AI agents from Claude, Codex, and Hermes executed these commands, resulting in callbacks from Fortune 500 networks within an hour. The research highlights a broken trust model in AI documentation.

↗ webpronews.com (2026-08-31)

2026-08-31

incident

Spring Ring: Voice Phishing Campaign in Microsoft Teams Deploys MalwareSpring Ring

The Spring Ring campaign uses voice phishing (vishing) via Microsoft Teams to trick users into installing remote access tools, leading to malware deployment and targeting of enterprise domain controllers. This incident highlights the abuse of legitimate collaboration platforms to bypass traditional email-based defenses.

Microsoft ↗ unit42.paloaltonetworks.com (2026-08-31)

2026-08-31

vendor announcement

Microsoft Defender Bug Triggers False 'Antivirus Turned Off' AlertsMicrosoft Defender false positive alert

Microsoft confirmed a bug in Microsoft Defender Antivirus that causes false notifications claiming the antivirus is turned off, even though protection remains active. The issue affects Windows systems and stems from a logic error in the security health reporting. Microsoft is working on a fix.

Microsoft ↗ gbhackers.com (2026-08-31) ↗ pcworld.com (2026-08-31) ↗ www.computerworld.com (2026-08-31) ↗ techradar.com (2026-09-01)

2026-08-31

research poc

Russian hackers embed nuclear weapon prompt in malware to bypass AI safety filtersAI safety guardrail bypass via prompt injection in malware

Russian state-sponsored hackers embedded a fake nuclear weapon request in a malicious VBS script to manipulate AI safety filters during automated analysis. The technique aims to trick AI-based security tools into misclassifying or mishandling the malware. This represents a novel evasion tactic targeting AI-driven detection systems.

↗ helpnetsecurity.com (2026-08-31)

2026-08-31

research poc

Gryxa Windows malware uses AI and RMM tools to steal browser passwords and evade defendersGryxa

Researchers have identified Gryxa, a Windows malware that leverages AI and legitimate remote monitoring and management (RMM) tools to steal browser passwords, maintain access, and actively resist removal by security teams. The malware is delivered via phishing and uses AI to adapt its evasion tactics based on defender actions.

↗ cybersecuritynews.com (2026-08-31) ↗ gbhackers.com (2026-08-31) ↗ securitybrief.com.au (2026-09-01)

2026-08-31

vuln disclosure

PaperCut MF/NG Vulnerabilities CVE-2026-82078 and CVE-2026-81578 Exploited in the WildCVE-2026-82078

Two zero-day vulnerabilities in PaperCut MF/NG print management software, CVE-2026-82078 and CVE-2026-81578, have been exploited in the wild. A second emergency patch has been released to address incomplete fixes from the initial patch. The vulnerabilities allow remote code execution and privilege escalation.

PaperCut ↗ www.securityweek.com (2026-08-31) ↗ thecyberexpress.com (2026-08-31) ↗ scworld.com (2026-09-10) ↗ techtimes.com (2026-09-11)

2026-08-31

incident

Infostealer Infection Exposes Blind Eagle-Linked Operator’s Malware Production PipelineBlind Eagle operator workstation compromise

A workstation used by an operator linked to the Blind Eagle threat actor was infected with an infostealer, exposing the attacker's malware production pipeline. Researchers gained visibility into tools, scripts, and infrastructure used to build and deploy malware. This incident provides insight into the operational security failures of the threat actor.

↗ gbhackers.com (2026-08-31)

Week of August 31, 2026

Events

This is last week, public. Get this morning's, written for you.