All news with #defense evasion tag

🛡️ Sophos analyzed a week of June 2026 telemetry and found AI coding agents like Claude Code, Cursor, and OpenAI Codex frequently trigger behavioral detection rules designed to catch human attackers. The agents perform actions—decrypting browser credentials, enumerating Windows Credential Manager, downloading files via LOLBins, and writing startup scripts—that look like malicious behavior to endpoint engines. While often benign developer automation, these behaviors overlap precisely with attacker techniques and can generate false positives. Sophos recommends scoping rules to agent parents, workspaces, and download reputations while keeping credential access tightly controlled.

🔍 The Gentlemen ransomware highlights challenges for CISOs in stopping attackers after an initial foothold. Researchers report the malware self-propagates using legitimate Windows management tools while attempting to disable security and recovery systems. Picus Security notes the encryptor, written in Go and obfuscated with Garble, leverages multiple lateral-movement methods and targets backups, EDR, and virtualization services to hinder recovery.

🖥️ Researchers at Shandong University demonstrated TrojPix, a novel covert channel that modulates otherwise imperceptible on-screen pixels so the video cable emits a faint radio signal a nearby receiver can decode. The technique requires only user-level malware that can draw to the screen and achieved a peak throughput of 8.1 Mbps and a laboratory range reported up to 208 meters. TrojPix works without hardware changes, can hide transmissions under normal-looking content or a fake powered-off display, and was tested across multiple monitor brands and cable types.

🛡️ Opera has added Paste Protect, a feature that intercepts and blocks ClickFix-style attacks which trick users into copying and running malicious commands. The mechanism builds on existing Hijack protection and a new Injection protection to detect and prevent harmful content from reaching the browser clipboard across Windows, macOS, and Linux. When suspicious content is blocked, Opera shows a warning, a red indicator in the address bar, and permits viewing the first 120 characters or approving the copy after a 5-second delay. The feature is enabled by default and can be managed via Settings → Privacy & Security → Paste Protect.

🔒 Analysis by ReliaQuest shows the ClickFix social engineering technique dominated malware delivery from March to May 2026. ClickFix tricks users into pasting attacker-supplied commands into trusted dialogs like Run, Terminal, or Script Editor, allowing payloads such as infostealers to execute while evading many defenses. The method has been used to deliver Windows malware and, notably, to deploy AMOS/Atomic Stealer to macOS via Script Editor. ReliaQuest urges equal monitoring for macOS and recommends user training and administrative restrictions to mitigate ClickFix risks.

🛡️ The ClickFix technique tricks users into executing malicious commands themselves by presenting convincing prompts like fake CAPTCHAs, Cloudflare checks, or “browser update” notices. Attackers rely on clipboard copy and instruct victims to paste commands into the Windows Run dialog, bypassing endpoint defenses that see the activity as legitimate user action. Check Point’s ThreatCloud AI team developed the ClickFix Engine, integrated into Gateways, Email Security, and Browse Security, to detect behavioral signals in page HTML and block such attacks irrespective of domain reputation.

🛡️ A malware developer has begun embedding provocative text about nuclear and biological weapons inside large JavaScript block comments in spyware payloads to confuse AI-based scanners. The commented header is ignored at runtime but aims to trigger refusals or misclassification in naive LLM-powered triage systems that ingest file starts without isolating untrusted content. Traditional detection methods—YARA, entropy checks, AST parsing, and behavioral analysis—remain effective, but the technique is a practical anti-analysis tactic against weak AI-first pipelines.

🔒 GitHub is updating the official actions/checkout action to refuse common pwn request patterns by default, effective June 18, 2026, with backports planned for July 16, 2026. The change prevents checking out forked pull request head or merge commits in pull_request_target and certain workflow_run events unless authors explicitly set allow-unsafe-pr-checkout to true. This aims to reduce attacks that exploit privileged workflows to steal secrets or the GITHUB_TOKEN.

🔍 Microsoft DART describes a complex multi-stage intrusion where two unrelated threat actors operated simultaneously, blending ransomware tactics with stealthy reconnaissance and persistence. Investigators observed exploitation attempts against on-premises SharePoint, use of legitimate tools like Velociraptor, cloud tunneling, credential misuse, and DLL sideloading to maintain access and evade detection. Coordinated telemetry correlation and threat intelligence enabled containment and targeted remediation guidance.

📰 This week’s recap highlights recurring attack patterns: abused integrations, poisoned websites, fake tools, and ransomware groups disabling security products. Notable incidents include the large-scale FortiBleed campaign compromising FortiGate devices, the Gentlemen RaaS developing the GentleKiller EDR-killing suite, and active exploitation of a critical Splunk flaw. Mobile and crypto-related malware campaigns also featured prominently.

🔍 New research from ESET reveals that The Gentlemen ransomware-as-a-service platform now supplies affiliates with an advanced EDR killer framework called GentleKiller, alongside third-party tools like HexKiller, ThrottleBlood and HavocKiller. The leak shows affiliates can deploy bring-your-own vulnerable driver (BYOVD) techniques to gain kernel privileges and disable hundreds of EDR processes across many vendors. ESET warns this lowers the bar for less skilled attackers and urges organizations to enforce protections such as HVCI and KMCI, apply strict driver allow/block policies, and regularly audit drivers.

🛡️ At least one malware author is inserting large comment blocks with policy-triggering content about nuclear and biological weapons into JavaScript payloads to disrupt AI-driven analysis. The decoy text sits inside comments so execution is unchanged while early-stage LLM-based triage can be confused or refuse to process the file. Traditional detection methods like YARA rules, entropy checks, and deobfuscation remain effective. This tactic targets naive pipelines that expose untrusted file starts to language models.

🛡️ Multiple ClickFix campaigns have been linked to three distinct loaders — BabaDeda Loader, Lorem Ipsum Loader, and Potemkin — delivering information stealers, backdoors, RATs, and other payloads against diverse sectors. The attacks rely on social-engineered ClickFix lures that trick victims into running PowerShell or command sequences, then use staged techniques such as hidden PowerShell, DLL side-loading, in-memory shellcode, and external payload storage to evade detection. Researchers from Morphisec, BlueVoyant, and Huntress attribute the campaigns to evolving, modular loader frameworks that separate delivery, storage, execution, and payload deployment for greater stealth.

🛡️ Attackers abuse NTFS junctions to create recursive directory loops that generate effectively infinite file paths, causing recursive scans and EDR products to hang. With only write access, an attacker can create junctions that point back to parent folders, producing GhostBranch or the more expansive GhostTree structures. These loops multiply possible paths exponentially, preventing file scanners from reaching malicious files and enabling evasion. Microsoft was notified and later patched the issue.

🛡️ Fortinet researchers uncovered a campaign where threat actors disguise malware as AI study guides and developer resources to deliver a multi-stage attack culminating in the AsyncRAT trojan. The booby-trapped archives contain shortcut (LNK) files and hidden documents that trigger staged scripts, using trusted system tools and AutoHotkey repurposed as an execution engine to evade detection. Attackers deploy scheduled tasks disguised as Realtek services, process hollowing to run payloads inside legitimate .NET processes, and hide components in decoy files to keep victims unaware while PowerShell stages execute silently.

🛡️ The Miasma credential-stealing worm, an evolution of the Shai-Hulud toolkit, was briefly published on GitHub after threat actors uploaded it to multiple compromised accounts. The framework steals developer build and cloud credentials, compromises package registries and repositories, and propagates autonomously without C2 by abusing GitHub. Researchers note destructive 'dead-man switch' behavior and a build pipeline that randomizes payloads to evade detection, increasing supply-chain risk.

🔍 Cloud logging services like AWS CloudTrail and Google Cloud Logging offer essential visibility into cloud activity but are also high-value targets for attackers. This article examines two primary attack goals—defense evasion and establishing continuous visibility—and demonstrates methods attackers use to disrupt or exfiltrate logs. It outlines practical attack techniques such as stopping logging, deleting storage or routers, abusing encryption keys, and log poisoning, and highlights detection and mitigation approaches.

🔍 Sophos X-Ops uncovered a lab where a threat actor used AI coding tools to develop and test malware aimed at evading EDR products. The files and Git repository showed Python scripts—many partially AI-generated—used to build and iterate evasion modules against vendors including Sophos, CrowdStrike and Microsoft. Humans retained control of the workflow, using AI to accelerate building, testing and refinement while operating inside an AI-native environment.

🛡️ This Microsoft Threat Intelligence blog dissects The Gentlemen, a Go-based RaaS that combines per-file ephemeral Curve25519/XChaCha20 encryption with aggressive self-propagation across networks. The post details operator models, command-line controls, speed modes, privilege elevation via scheduled tasks, and extensive defense-evasion steps including disabling Defender, deleting shadow copies, clearing logs, and terminating backup, database, virtualization, and EDR services. Practical mitigations, Defender detections, hunting queries, and IOCs are provided for defenders and incident responders.

🛡️ Microsoft is previewing an automatic device isolation feature in Defender for Endpoint to help contain active cyberattacks by severing most network traffic while preserving connections to security services. The capability is part of its auto attack disruption tool within Defender XDR, and Microsoft says actions are time-limited and can be tuned or reversed by administrators. A new SANS Institute paper warns threshold-driven autonomous containment can be weaponized to disable user accounts, underscoring the need for careful configuration and governance.