Koffei

Daily CyberNews · 27/10/2026

‹ 26/10/202628/10/2026 ›
Today in 30 seconds
  • 🔴 Patch PHP to 8.5.12, 8.4.27, 8.3.36 or 8.2.35: PHP2Shell is exploited and in KEV.
  • 🔴 Block external Teams "IT Support" chats: STAC4749 vishing leads to Chaos ransomware.
  • 🔴 Audit GitHub for ci-sync- branches: Gheist steals CI/CD secrets with stolen tokens.
  • 🔴 Patch Magento/Adobe Commerce: 953 stores skimmed via CVE-2026-48356 and CVE-2025-54236.
  • 🟠 Patch Brotli everywhere for CVE-2026-111497, including the new .NET out-of-band updates.

Vulnerabilities 5

🔴 Exploited or active

PHP2Shell flaws now confirmed actively exploited, added to CISA KEV

CVE-2026-109618CVE-2026-109619in KEV

WhatThe PHP2Shell unauthenticated RCE flaws are confirmed exploited and in KEV, with a leaked exploit tool circulating and 32.5M+ exposed PHP services unpatched.

DoUpgrade to PHP 8.5.12, 8.4.27, 8.3.36 or 8.2.35; federal deadline is Oct 29.

Details and sources
CVE-2026-109618 and CVE-2026-109619 (both CVSS v4.0 9.2 / v3 9.8, unauthenticated RCE affecting PHP-FPM, Apache mod_php, php-cgi, FrankenPHP and LiteSpeed; fixed in PHP 8.5.12/8.4.27/8.3.36/8.2.35) were added to CISA's Known Exploited Vulnerabilities catalog on Oct 26, giving federal agencies until Oct 29 to patch. Perplexity researcher Adel Karimi confirmed working RCE with memory disclosure against honeypots on Oct 24, and a leaked exploit tool nicknamed "PHP2Shell" is circulating; Censys estimates 32.5M+ exposed PHP services remain unpatched. The Hacker News (source)
🔴 Exploited or active

Suspected North Korea-linked scanning wave against PHP2Shell flaws

PHP2ShellLazarus Groupscanning

WhatCYFIRMA links a low-and-slow scan of 216 unpatched PHP2Shell hosts in 12 countries to a Lazarus-ecosystem operator, single-source and unconfirmed.

DoFully patch PHP on government, finance, logistics, aviation and manufacturing hosts.

Details and sources
CYFIRMA assesses with "moderate confidence" that a Korean-speaking operator tied to the Lazarus Group ecosystem is running a deliberately low-and-slow scan (21 nodes, closed I2P coordination channel) against 216 unpatched/partially-mitigated CVE-2026-109618/CVE-2026-109619 hosts across 12 countries (government, finance, logistics, aviation, manufacturing sectors). Flagged as single-source/unconfirmed — Palo Alto Unit 42 separately retracted an earlier claim linking a European government breach to the same flaws. The Hacker News (source)
🟠 High impact

Critical Brotli decompression flaw forces emergency patches across .NET, Node.js and PHP

CVE-2026-111497CVSS 9.2Brotli

WhatA buffer overflow in Google's Brotli decoder 1.0.3 to 1.2.0 risks crashes, memory disclosure or RCE; PoCs exist but no exploitation is confirmed.

DoUpdate Brotli to 1.0.10, 1.1.1 or 1.2.1, Node.js to 22.23.4, 24.22.1 or 26.12.1, and apply the .NET 8/9/10 out-of-band updates.

Details and sources
CVE-2026-111497 (CVSS v4.0 9.2 / v3 9.8; CWE-787 out-of-bounds write plus CWE-252) is a buffer overflow in Google's Brotli streaming decoder (versions 1.0.3–1.2.0, fixed in 1.0.10/1.1.1/1.2.1) that can be triggered by a crafted compressed stream, risking crashes, memory disclosure or RCE. Oct 26 brought a fresh wave of emergency releases — Microsoft shipped out-of-band .NET 8/9/10 (and .NET 11 RC3) updates and Node.js shipped 22.23.4/24.22.1/26.12.1 — on top of PHP's own fix. PoCs exist on GitHub (found by OpenAI's red-teaming agents) but no confirmed in-the-wild exploitation yet (EPSS ~36%). BleepingComputer (source) · The Hacker News (source)
🟠 High impact

Two critical Wasmtime/Cranelift flaws allow WebAssembly sandbox escape

CVE-2026-115889CVE-2026-115893CVSS 9.9Wasmtime

WhatTwo Cranelift compiler bugs could let a malicious WebAssembly module escape the Wasmtime sandbox; no exploitation is confirmed.

DoUpdate Wasmtime to 36.0.19 or a later maintenance release; 24.x LTS stays unpatched.

Details and sources
CVE-2026-115889 (all architectures, CVSS v3.1 9.9) and CVE-2026-115893 (ARM64-only, present since 2020) are bugs in Wasmtime's Cranelift compiler backend — a 24-bit SSA value-index overflow and an out-of-range branch-table bug, respectively — that can let a malicious WebAssembly module escape the sandbox via miscompilation/type confusion. Fixed in Wasmtime 36.0.19 and later maintenance releases (Oct 19–20); the unsupported 24.x LTS line stays unpatched. No confirmed in-the-wild exploitation. The Hacker News (source)
🟠 High impact

Microsoft Authenticator flaw let permission-less Android apps steal work-account tokens

Added after publishingCVE-2026-80097CVSS 8.6Microsoft Authenticator

WhatA permissionless Android app could impersonate Teams via Authenticator's AccountChooserActivity and obtain work-account sign-in tokens.

DoUpdate Microsoft Authenticator for Android to 6.2606.4246.

Details and sources
CVE-2026-80097 (CVSS 8.6) in Microsoft Authenticator for Android let a malicious app with zero permissions impersonate a trusted app like Teams and obtain sign-in tokens for a victim's signed-in work account, granting access to mail, files and calendar in researchers' tests. The flaw sat in an exposed AccountChooserActivity component: Authenticator checked the calling app's signature against the real Teams signature but didn't use the result to gate the token. Disclosed by researcher Ofek Levin (Enclave fellowship) on Oct 23; fixed in Authenticator 6.2606.4246 as part of September 2026 Patch Tuesday. Microsoft rates exploitation "less likely" and says it requires a malicious app already on-device plus a completed Authenticator sign-in flow — disputing Enclave's claim that no further user interaction is needed. Not listed in CISA's KEV catalog as of Oct 26; unclear whether updating revokes tokens a malicious app may have already obtained. The Hacker News (source)

Breaches / Incidents 3

🟠 High impact

University of Copenhagen data breach exposes 525,000 people

University of Copenhagen525,000 peopleCPR numbers

WhatHackers breached a university test environment, exposing names, Danish CPR numbers and contacts of 525,000 people plus access PINs for about 20,000.

Details and sources
Hackers breached one of the university's test environments; names, Danish national ID (CPR) numbers, phone numbers and emails were exposed, plus physical-access PIN codes for an additional ~20,000 staff/students. The university restricted physical access to some campus locations as a precaution and reported the breach to the Danish Data Protection Agency; no attacker or cause has been disclosed. BleepingComputer (source)
🟡 Worth knowing

DreamHost forces password resets amid unconfirmed ongoing incident

DreamHostforced password resets

WhatDreamHost has forced email and FTP/SFTP password resets in an unconfirmed ongoing incident since Oct 23.

DoDreamHost users should check email-forwarding rules for unrecognized destinations.

Details and sources
The US hosting/registrar has been dealing with an incident since Oct 23 that began as File Manager outages and escalated into forced password resets for email and FTP/SFTP accounts; File Manager was still down as of Oct 26. DreamHost has not confirmed a breach or root cause, though it has told users to check email-forwarding rules for unrecognized destinations — suggestive that some mailboxes were compromised — and did not respond to requests for an affected-account count. BleepingComputer (source)
🟡 Worth knowing

Keystone to destroy 18,835 hardware wallets after attacker steals device-verification key

Added after publishingKeystonehardware walletsverification key

WhatAn attacker stole Keystone's device-verification key and about 26,000 buyer records, so Keystone will destroy 18,835 unsold wallets.

Details and sources
Keystone disclosed that an attacker who breached its website (via a vulnerability in an unnamed third-party data-monitoring tool) obtained the cryptographic key used to verify its hardware wallets are genuine; access was found Oct 20, blocked, and Keystone says the underlying flaw is fixed. Anyone holding the key could build counterfeit devices that pass Keystone's online verification check — though the key doesn't protect recovery phrases, private keys or transaction signing, and Keystone reports no counterfeits or thefts so far. Keystone will destroy all 18,835 unsold devices carrying the exposed key and re-key future production; already-shipped devices can't be re-keyed since the key is burned in at manufacture. The breach also exposed 25,964 buyer records (names, contact/shipping details, orders) plus 12,873 unpaid-checkout records from its Shopify-hosted store — no payment, bank or government-ID data, per Keystone — and the company says its own bulk-deletion tooling for old shipping data failed silently, which is why some of that data was still present to be exposed. The Hacker News (source)

Malware / Ransomware 8

🔴 Exploited or active

Microsoft Teams "IT Support" vishing campaign deploys Chaos ransomware

STAC4749GoGRPCChaos ransomwareTeams

WhatSTAC4749 mail-bombs targets, then poses as IT support in Teams external chat to get RMM access, deploying GoGRPC and Chaos ransomware.

DoRestrict Teams external chat and hunt for unexpected Quick Assist, AnyDesk, ScreenConnect or TeamViewer sessions after mail-bombing.

Details and sources
A cluster tracked as STAC4749 mail-bombs targets, then uses a spoofed "IT Support"/"Help Desk" Teams external chat to talk victims into granting remote access via Quick Assist, AnyDesk, ScreenConnect, TeamViewer and similar RMM tools. Post-access activity includes AD reconnaissance, a new GoGRPC backdoor, lateral movement to domain controllers, data exfiltration and ultimately Chaos ransomware deployment; the group avoids pasting URLs in Teams to dodge Safe Links scanning. The Hacker News (source)
🔴 Exploited or active

"Gheist" campaign weaponizes stolen GitHub tokens to exfiltrate CI/CD secrets

GheistGitHub tokensCI/CD secrets

WhatAn active campaign uses stolen GitHub tokens to push ci-sync-* branches that dump secrets into Actions logs, affecting 850+ accounts.

DoSearch for ci-sync-* branches and quickly deleted Actions runs, then rotate CI/CD secrets.

Details and sources
Tracked by Wiz since mid-September and still active, Gheist uses stolen GitHub write-access tokens (sourced from infostealer logs and prior supply-chain leaks) to push a ci-sync-* branch whose Actions workflow base32-encodes every readable secret into the job log, pulls the log via the API, then deletes the run within seconds — leaving no outbound C2 traffic. Over 850 GitHub accounts and 13,500+ planted branches have been found so far, including enterprise private repos. The Hacker News (source) · Dark Reading (source)
🟡 Worth knowing

Malicious AGENTS.md file hijacks AI coding agents from the Downloads folder

AGENTS.mdCLAUDE.mdAI coding agentsPoC

WhatA planted instruction file in Downloads is auto-loaded by AI coding agents run there, tricking them into installing a rogue npm package.

DoDo not run AI coding agents from the Downloads folder.

Details and sources
Forever Security found a planted AGENTS.md/CLAUDE.md-style instruction file in a victim's Downloads folder that, when an AI coding agent (Claude Code, OpenAI Codex, Cursor CLI) is run from that directory, is auto-loaded and tricks the agent into installing a rogue npm package. A related technique against Claude Code redirected its API traffic to leak an OAuth token, achieving code execution in roughly half of tests; a harmless-payload test against Codex succeeded about 90% of the time. No attacker attribution — this is a disclosed technique/PoC, but researchers note a malicious webpage could plant the file with no user interaction. BleepingComputer (source) · Dark Reading (source)
🔴 Exploited or active

953 online stores infected with scanner-evading JavaScript skimmer

Magento skimmerCVE-2026-48356CVE-2025-54236

WhatA scanner-evading JavaScript skimmer hidden in cms_block entries infected 953 Magento/Adobe Commerce stores, mainly via PolyShell and SessionReaper.

DoCheck Magento cms_block entries for injected loaders.

Details and sources
Sansec reports a campaign, active since mid-2025 and escalating this month, injecting malicious loaders into Magento/Adobe Commerce cms_block entries; the loader fingerprints visitors and serves its payload only to real shoppers, not security scanners. Initial access came mainly through the unauthenticated file-upload flaw CVE-2026-48356 ("PolyShell," ~471 stores, no isolated patch yet for current production versions) and CVE-2025-54236 ("SessionReaper"). The Hacker News (source)
🟠 High impact

"uCreator" rentable Android spyware hijacks Telegram and WeChat accounts

uCreatorAndroid spywareTelegramWeChat

WhatuCreator spreads through tampered Telegram and WeChat clones, hijacking Telegram accounts via 2FA interception and cloning WeChat sessions.

DoInstall Telegram and WeChat only from official stores.

Details and sources
Distributed via tampered Telegram/Telegram X and WeChat app clones sideloaded through fake app-store listings, uCreator intercepts Telegram's 2FA pre-hash to hijack or lock out accounts and virtualizes WeChat to clone login sessions for bulk friend-adding and account resale. Tentatively (unproven) linked to three companies in Hunan, China; researchers connect its Telegram component to the "Baohuo" backdoor first reported in 2025. The Hacker News (source)
🔴 Exploited or active

Microsoft 365 "token spray-and-replay" campaign hits 97 accounts at 79 organizations

Microsoft 365AitM token replayStarkRDP

WhatAn AitM campaign replayed harvested M365 tokens to compromise 97 accounts at 79 organizations, registering rogue devices and stripping MFA.

DoReview new Entra devices and authenticator apps and look for sign-ins from StarkRDP/rdp.sh infrastructure.

Details and sources
Huntress disclosed a months-long adversary-in-the-middle campaign that harvests M365 sign-in tokens via phishing relays and replays them en masse from attacker-controlled devices (497 successful logins out of 200,000+ attempts), registering rogue Entra devices/authenticator apps for persistence and stripping MFA at 19+ organizations. Infrastructure traces to a German host tied to StarkRDP/rdp.sh. The Hacker News (source)
🟡 Worth knowing

Lazarus abuses South Korean security software in "Operation FlashHole" espionage campaign, deploys malware with post-quantum encryption

Added after publishingLazarus GroupFlashHoleCubridgeCOPPERHEDGE

WhatLazarus targets South Korean firms via fake résumé pages and INISAFE CrossWeb EX abuse, delivering Cubridge with ML-KEM-1024 C2.

DoHunt for DLL sideloading via fsquirt.exe; LPEClient was also paired with Windows zero-day CVE-2026-68820.

Details and sources
Kaspersky attributes Operation FlashHole to the North Korea-linked Lazarus Group (a continuation of 2025's Operation SyncHole), active since at least July 2025 against at least three South Korean construction, software and wire-communication-equipment manufacturers. Spear-phishing links to fabricated résumé/portfolio pages (signs point to GitHub Pages hosting) lead to DLL sideloading via fsquirt.exe that chains through TechHub, TouchHub, TOUCHSHIFT, the previously-undocumented ClientHub, and LPEClient to deliver the Cubridge orchestration framework; the campaign abuses Initech's mandatory INISAFE CrossWeb EX financial security software, which Lazarus has abused before in 2021–2023 watering-hole attacks. The latest malware encrypts its C2 channel with ML-KEM-1024, a post-quantum algorithm Kaspersky traces back to September 2024. LPEClient was separately seen paired with a Windows zero-day, CVE-2026-68820 (reported by Check Point, Aug 2026); concurrent COPPERHEDGE backdoor activity against South Korean targets posted stolen data to a dark-web leak site named ULOSE. The Hacker News (source)
⚪ Context

AI-generated fake login pages are making account takeovers easier

Added after publishingAI phishing pagesAiTMexplainer

WhatA vendor trend piece says generative AI makes cloned login pages and AiTM token theft easier.

DoUse phishing-resistant MFA or passkeys and monitor authentication and session activity.

Details and sources
A trend piece (contributed by Keeper Security) notes generative AI now lets attackers rapidly clone legitimate login screens and tailor phishing to specific employees via lookalike domains and personalized IT-style notifications; more advanced adversary-in-the-middle (AiTM) setups proxy the real authentication flow in real time to steal session cookies/tokens, which can bypass MFA entirely. Citing Verizon's 2026 DBIR, phishing accounted for 44% of the AI-assisted initial-access techniques it examined. No specific incident, victim count, CVE or threat actor is named — recommended mitigations cited include phishing-resistant MFA/passkeys, browser-level phishing protection, and monitoring of auth/session activity. The Hacker News (source)

Threat Actor Activity 3

🟡 Worth knowing

South Korean "AI pentester" hacking gang arrested — 8.2 million records stolen

South KoreaARTEX AIarrest

WhatPolice arrested a former pentester whose seven-person gang used ARTEX AI to hit 90+ South Korean organizations and steal 8.2 million records.

Details and sources
A 32-year-old former penetration tester led a seven-person gang that hit 90+ South Korean organizations (schools, hospitals, corporations) since September 2025, escalating in late August using an autonomous AI pentesting framework ("ARTEX AI") to automate intrusions; the same tool has separately been tied to attacks on South Korean banks and, per Microsoft, to Chinese state-linked activity. The gang demanded roughly 360 million won (~$270,000) per ransom in 38 incidents. The leader was arrested Oct 15 and formally detained Oct 17; six co-conspirators remain under investigation. Billed by Korean police as the country's first arrest for large-scale AI-driven hacking. BleepingComputer (source)
🟡 Worth knowing

Chinese national extradited over $36.9M "pig butchering" crypto scam

Pig butcheringextradition$36.9M

WhatA Chinese national extradited from the Netherlands was arraigned over a $36.9M crypto investment scam with 174 US victims.

Details and sources
Zhen Zhao, extradited from the Netherlands, was arraigned in California on a money-laundering conspiracy charge tied to a scheme that lured 174 US victims via social media and dating apps into fake crypto investments, routing proceeds through a Bahamas bank account. He faces up to 20 years; eight co-conspirators have already pleaded guilty. BleepingComputer (source) · DOJ press release (source)
🟡 Worth knowing

Australia disconnects Amadey malware operation, cutting off ~180,000 infected machines

Added after publishingASDAmadeyStealCdisruption

WhatAustralia's ASD cut an Amadey backend off from about 180,000 infected machines and destroyed 5TB of stolen data from a criminal marketplace.

Details and sources
The Australian Signals Directorate (ASD) says it gained access to the backend of an Amadey malware-as-a-service operation and severed the controller from roughly 180,000 infected computers worldwide, including some in Australia; Amadey has been sold as a crime kit since 2018 (ESET counts 53 separate botnets with no shared infrastructure) and is mainly used to install further malware such as the StealC infostealer. The action is separate from Europol's June 2026 Operation Endgame takedown of Amadey/StealC/SocGholish infrastructure, which did not list Australia or ASD. ASD separately disclosed it broke into a criminal marketplace and destroyed 5TB of stolen data — including an estimated 4.8 million records on Australians — before it could be sold. The Hacker News (source)
‹ 26/10/202628/10/2026 ›