Koffei

Daily CyberNews · 23/10/2026

‹ 22/10/202626/10/2026 ›
Today in 30 seconds
  • 🔴 Patch Plex Media Server to 1.43.3.10861: Plex2Shell is exploited and in KEV.
  • 🔴 Roll LiteLLM back to 1.106.2 and rotate credentials if v1.107.0 ran anywhere.
  • 🔴 Patch PHP now: CVE-2026-109618 and CVE-2026-109619 face millions of exploit attempts.
  • 🔴 ConnectWise Cloud Backup users: hunt for Storm-0940 password spraying and Graph privilege changes.
  • 🟠 Upgrade Node.js to 22.23.4, 24.22.1 or 26.12.1 for CVE-2026-111497.

Vulnerabilities 12

🟠 High impact

GitLab critical unauthenticated RCE

CVE-2026-110264CVSS 10.0GitLab

WhatAn integer overflow in GitLab's bundled Onigmo regex engine allows unauthenticated RCE on self-managed GitLab 7.0 and later; no exploitation is reported.

DoUpgrade self-managed GitLab to the October 21 fixed release for your 18.11 to 19.5 branch.

Details and sources
CVE-2026-110264 (CVSS 10.0) is an integer overflow in the bundled Onigmo regex engine, triggerable by a specially crafted regular expression, that can allow unauthenticated remote code execution on self-managed GitLab 7.0 and later (GitLab.com/Dedicated unaffected). Fixed across the 18.11–19.5 branches on October 21; GitLab reports no sign of exploitation and it is not yet CISA KEV-listed. The Hacker News (source)
🔴 Exploited or active

Plex2Shell chain actively exploited

Plex2ShellCVE-2026-96656in KEV

WhatFour chained Plex Media Server flaws give unauthenticated RCE and are used to deploy cryptominers and proxy malware.

DoUpdate Plex Media Server to 1.43.3.10861; federal deadlines are Oct 23 and Nov 3.

Details and sources
Four chainable Plex Media Server flaws CVE-2026-96652 (5.3, SSRF), CVE-2026-96655 (5.3, SSRF), CVE-2026-96654 (6.9, encoded-URI-scheme handling) and CVE-2026-96656 (8.6, path/filename control) — can be combined for unauthenticated RCE and are being actively used to deploy cryptominers and proxy malware. Fixed in Plex Media Server 1.43.3.10861; CISA added all four to its KEV catalog on October 20 with federal patch deadlines of October 23 (CVE-2026-96656/CVE-2026-96654) and November 3 (CVE-2026-96652/CVE-2026-96655). The Hacker News (source) · BleepingComputer (source)
🟠 High impact

Node2Shell — Node.js Brotli decoder RCE

CVE-2026-111497CVSS 8.1Node2Shell

WhatAn out-of-bounds write in Node's bundled Brotli decoder can lead to RCE, with a working PoC against Node.js 24.22.0 but no exploitation.

DoUpgrade to Node.js 22.23.4, 24.22.1 or 26.12.1; move off EOL 20.x and earlier.

Details and sources
CVE-2026-111497 (CVSS 8.1) is an out-of-bounds write in Node's bundled Brotli decoder, triggered by a crafted Brotli stream hitting an impossible block-type state, that can lead to remote code execution; Imperva demonstrated a working PoC against a stock Node.js 24.22.0 build. Fixed in Node.js 22.23.4, 24.22.1 and 26.12.1 (20.x and earlier are EOL and won't get a fix); no in-the-wild exploitation reported as of October 22. The Hacker News (source)
🟠 High impact

ShadowForge — NVIDIA AIStore privileged-container escape

CVE-2026-65162CVSS 10.0NVIDIA AIStore

WhatAIStore's legacy ETL API accepts dangerous Kubernetes settings, allowing privileged container creation and possible host takeover; no PoC or exploitation is confirmed.

DoUpgrade NVIDIA AIStore to 5.0.

Details and sources
CVE-2026-65162 (CVSS 10.0) lets AIStore's legacy ETL API accept dangerous Kubernetes host/security settings, enabling privileged container creation and potential host takeover; disclosed under the name "ShadowForge" by AegisSec Research on October 22, though NVIDIA had already shipped the fix (AIStore 5.0) on October 13 alongside five related, lower-severity AIStore flaws (CVE-2026-65173, CVE-2026-65172, CVE-2026-65170, CVE-2026-65101, CVE-2026-65164/CVE-2026-65103). No public PoC or in-the-wild exploitation confirmed. The Hacker News (source)
🟠 High impact

Laravel TokenGuard authentication bypass

CVE-2026-114509CVSS 9.2Laravel TokenGuard

WhatOn some MySQL/MariaDB setups, Laravel TokenGuard accepts a crafted non-string credential, letting an unauthenticated attacker take over existing accounts.

DoUpgrade Laravel to 12.68.0 or 13.29.0.

Details and sources
CVE-2026-114509 (CVSS 9.2) lets an unauthenticated attacker authenticate as another existing user and take over their account on TokenGuard-protected routes, on apps using certain MySQL/MariaDB setups where TokenGuard fails to reject a crafted non-string credential. Affects Laravel below 12.68.0 and 13.0.0–13.28.x; fixed in 12.68.0 and 13.29.0; no exploitation reported. The Hacker News (source)
🟠 High impact

Unpatched Linux KSMBD kernel flaws, leaked early

Linux KSMBDZDI-CAN-33549ZDI-CAN-33774CVSS 9.0

WhatTwo unpatched KSMBD kernel SMB server flaws were accidentally posted publicly; no CVE, no mainline fix and no exploitation yet.

Details and sources
Two CVSS-9.0 flaws in KSMBD, the Linux kernel's built-in SMB server — an integer underflow in the SMB Direct/RDMA path (ZDI-CAN-33549) and a handler-table race letting SMB1 data hit an SMB2 handler (ZDI-CAN-33774) — were publicly posted to the Linux kernel mailing list after Trend Micro's Zero Day Initiative accidentally sent them there instead of the private security list. No CVE has been assigned yet, no fix has landed in mainline (a fix for the RDMA bug is reportedly queued for the next cycle), and no in-the-wild exploitation is reported. The Hacker News (source) · Phoronix (source)
🟡 Worth knowing

eMudhra CA private-key weakness forces SSL/TLS exit

eMudhraCA exit30,800 certificates

WhatAfter weak intermediate keys were recovered, Chrome and Firefox distrusted six eMudhra intermediates and eMudhra is leaving the global SSL/TLS business.

DoReplace any certificates issued under the affected eMudhra intermediates, including those issued via InCommon.

Details and sources
A researcher recovered the private keys of eight eMudhra intermediate CA certificates on October 17 because the keys were far smaller than the 256-bit minimum needed to resist brute-forcing. Chrome and Firefox withdrew trust from six affected intermediates under eMudhra's root, disrupting roughly 30,800 certificates (mostly issued via InCommon). eMudhra says the keys were never used to mis-issue certificates and its systems weren't breached, but is now exiting the global SSL/TLS certificate business entirely, citing the business's low revenue share plus the compliance burden and scrutiny (including unanswered questions from Chrome over a 2025 revocation-list incident). The Hacker News (source)
🟠 High impact

Fortinet ships critical patches across FortiOS/FortiManager/FortiAnalyzer/FortiMail/FortiAuthenticator

FortinetFortiOSFortiManagerFortiMail

WhatFortinet fixed multiple critical flaws across five products without CVE IDs, warning that patch-diffing could yield exploits within hours to days.

DoUpgrade FortiOS to 7.2.14, 7.4.13, 7.6.8 or 8.0.2 and patch FortiManager, FortiAnalyzer, FortiMail and FortiAuthenticator.

Details and sources
Fortinet released fixes for multiple critical flaws spanning FortiOS, FortiManager, FortiAnalyzer, FortiMail and FortiAuthenticator (e.g. FortiOS 7.2–8.0 branches fixed in 7.2.14/7.4.13/7.6.8/8.0.2); Fortinet did not publish individual CVE IDs or CVSS scores and reports no active exploitation, but warns AI-assisted patch-diffing could produce working exploits within hours to days. Cyber Security News (source)
🟡 Worth knowing

QPACK amplification research against HTTP/3 servers

HTTP/3QPACK amplificationCVE-2026-12523

What7 of 10 HTTP/3 implementations let a small HEADERS frame amplify up to about 4000x in memory, exhausting default servers.

DoUpdate quic-go to 0.57.0 (CVE-2025-64702), quiche to 0.29.3 and Wireshark to 4.6.2 or 4.4.12 (CVE-2025-13945).

Details and sources
FoxIO researchers found that 7 of 10 tested HTTP/3 implementations mishandle QPACK header decompression, letting a single small HEADERS frame amplify into enough server-side memory to exhaust a default-configured server (amplification up to ~4000x); affected/fixed implementations include quic-go (CVE-2025-64702, fixed 0.57.0), Cloudflare quiche (CVE-2026-12523, fixed 0.29.3) and Wireshark's dissector (CVE-2025-13945, fixed 4.6.2/4.4.12), while Bun, the Rust h3 crate and aioquic/mitmproxy remain unpatched. Research disclosure only — no in-the-wild exploitation reported. FoxIO (source)
🟠 High impact

ASUS router unauthenticated RCE chain (no CVE yet)

ASUS routersRT-AX86U Propublic PoC

WhatA public PoC chains admin-token forgery with NTP-setting command injection for unauthenticated RCE on ASUS routers, with some models still unpatched.

DoInstall the latest ASUS router firmware where available.

Details and sources
Independent researcher mrbruh published a chain combining an admin-token forgery (an unregistered device_id skips the token check) with a command injection via the NTP server setting (an unescaped ; reaches nslookup), giving unauthenticated RCE on the RT-AX86U Pro and, by the author's estimate, many ASUS routers from 2018 onward. A public PoC exists on GitHub; ASUS has patched some affected models but not all, and no CVE has been assigned. No in-the-wild exploitation reported. mrbruh (source)
⚪ Context

PaperCut, Albert Heijn and the French supermarket refrigeration attack

PaperCutAlbert HeijnFrench refrigeration

WhatNo new developments on PaperCut, Albert Heijn or the French supermarket refrigeration intrusions.

Details and sources
checked today, nothing genuinely new beyond what's already tracked — no further patch bypasses or confirmed mass exploitation for PaperCut, no fresh official confirmation/attribution for Albert Heijn, and still no claimed responsibility for the French refrigeration intrusions.
🔴 Exploited or active

PHP2Shell update — mass exploitation attempts, still no confirmed compromise

PHP2ShellCVE-2026-109618CVE-2026-109619Potassium

WhatMillions of exploit attempts, mostly the Potassium Mirai variant, hit PHP servers; no successful RCE is confirmed, but VulnCheck showed RCE on default PHP-FPM.

DoPatch PHP; Censys counts over 34.4 million exposed unpatched PHP services.

Details and sources
Attempted exploitation of CVE-2026-109618/CVE-2026-109619 escalated sharply — Cloudflare logged millions of attempts by October 22, mostly a Mirai variant dubbed Potassium scanning and trying to recruit internet-facing PHP servers (Hunt.io: 123 targeted hosts) for DDoS against a single Ukrainian network. Multiple researchers (Previdian, Hunt.io, SOCRadar) still report no successful RCE or dropped webshells as of October 22–23, though VulnCheck has since demonstrated an ASLR-bypass technique that achieves RCE on default production PHP-FPM/x86-64 configurations, and Censys estimates over 34.4 million PHP services remain unpatched and exposed. One independent researcher (Elias Taylor) pushed back on the severity hype, arguing most public PoCs either depend on a pre-existing LFI flaw or quietly modify the target server to fake success, and that this is "not a Log4Shell-level event" — a useful counterpoint, though patching remains recommended regardless. The Hacker News (source) · Elias Taylor (source)

Breaches / Incidents 8

🔴 Exploited or active

LiteLLM release pipeline hacked to push info-stealer

LiteLLMv1.107.0PyPIDocker HubGHCR

WhatA malicious LiteLLM v1.107.0 ran on any Python invocation, stole cloud, AI and browser credentials, and installed an hourly loader.

DoRoll back to LiteLLM 1.106.2, look for cook.pth, cook.py and sitecustomize.py, block litellm[.]dev and rotate credentials.

Details and sources
An attacker compromised LiteLLM's release pipeline on October 21 and published a malicious v1.107.0 to PyPI, Docker Hub and GHCR for roughly two hours before it was pulled. The tainted build drops cook.pth/cook.py/sitecustomize.py, which run automatically on any Python/pip invocation, harvest cloud, AI-provider and developer credentials plus browser data, exfiltrate it encrypted to litellm[.]dev, and install a loader that fetches further code hourly. Recommended recovery is the last clean release, 1.106.2. BleepingComputer (source)
🟠 High impact

Japan's Onamae Desktop Cloud (GMO Internet) hacked

GMOOnamae Desktop Cloud26,700 VPS

WhatUp to about 26,700 customer VPS instances were infected with malware and GMO shut down the whole Desktop Cloud service.

Details and sources
Up to roughly 26,700 customer VPS instances were infected with malware; GMO Internet shut down the entire Desktop Cloud service with no restart timeline while it investigates. No actor or malware family has been named yet. BleepingComputer (source)
🟠 High impact

Railway PaaS breach exposes customer secrets

RailwayPaaSenvironment variables

WhatAn intruder accessed and decrypted Railway customer environment variables; drained crypto wallets are claimed but unconfirmed.

DoRailway customers should rotate any secrets stored in environment variables.

Details and sources
Railway disclosed that an intruder accessed and decrypted customer environment variables; a researcher claims some cryptocurrency wallets were drained, which Railway has not confirmed. Credentials have been rotated and service is partially restored. The Hacker News (source)
🟡 Worth knowing

City of Columbia, South Carolina hit by cyberattack

Columbia, South Carolinacity government

WhatA cyberattack took down City of Columbia systems including MyColaSC; SLED is investigating and ransomware is unconfirmed.

Details and sources
City government systems, including MyColaSC and GIS/animal-services platforms, went down; South Carolina's state law-enforcement division (SLED) is investigating. No actor has claimed responsibility and ransomware involvement is unconfirmed. The Record
🟡 Worth knowing

State Bank of India customer PII leaked

State Bank of IndiaBreachForums3,897 records

WhatLeaked SBI customer data including KYC records was posted on BreachForums, and SBI confirmed 3,897 leaked records.

Details and sources
An anonymous actor posted a batch of SBI customer data (names, mobile numbers, emails, addresses, dates of birth, bank profile IDs, loan metadata, and reportedly Central/video KYC records) on BreachForums; SBI disclosed the leak to India's stock exchange, confirmed 3,897 leaked customer records, and called the impact "a very, very small number of customers" (about 0.0008%). Cause is unclear — likely a peripheral system rather than core banking — and core operations are unaffected. Dark Reading (source)
🟡 Worth knowing

PingCAP breach

PingCAPstolen service-account credentials

WhatAttackers used stolen service-account credentials to copy PingCAP source code and customer records between Sept 9 and Oct 7.

Details and sources
Attackers used stolen service-account credentials to copy source code and customer records between September 9 and October 7, disclosed as part of The Hacker News' October 22 ThreatsDay roundup. The Hacker News (source)
🟡 Worth knowing

Zeabur shutting down after credential-theft breach

Zeaburfollow-upshutting down

WhatZeabur will shut down after leaked credentials were used to steal 3.2M environment-variable records and data on 321,810 accounts.

Details and sources
Hosting platform Zeabur announced it will shut down after an October attack used leaked credentials to steal over 3.2 million environment-variable records and account data for 321,810 accounts. The Hacker News (source)
⚪ Context

Quad9 to stop serving France over DNS-blocking order

Quad9FranceDNS-blocking order

WhatQuad9 will stop serving France on Nov 24 rather than comply with a court-ordered DNS block, ending its malware blocking for French users.

DoFrench organizations using Quad9 should plan a replacement protective DNS resolver before Nov 24.

Details and sources
A French court ordered DNS resolver Quad9 to block domains for broadcaster beIN's piracy dispute; Quad9 says it cannot identify French users without tracking them and will not comply, and will instead stop serving France entirely on November 24 — meaning French users (including government, finance, education and healthcare networks) lose Quad9's malware/phishing DNS-blocking protection. Quad9 (source)

Malware / Ransomware 2

🟡 Worth knowing

Lazarus Group's "Operation FlashHole" uses post-quantum cryptography

Lazarus GroupFlashHoleCubridgeSouth Korea

WhatLazarus uses ML-KEM-1024 to protect C2 for its Cubridge backdoor, luring South Korean firms via pages abusing INISAFE CrossWeb EX and fsquirt.exe.

DoHunt for in-memory shellcode launched via fsquirt.exe.

Details and sources
Kaspersky details a North Korean Lazarus Group campaign (part of its broader "Cascade Hole" activity, linked with medium-to-high confidence to the earlier "Operation SyncHole") that uses the NIST-approved ML-KEM-1024 post-quantum algorithm to protect command-and-control payload delivery for its modular Cubridge backdoor, alongside the LPEClient, TechHub and ClientHub loaders. Victims — at least three South Korean software, construction and telecom-equipment firms — are lured via phishing pages that abuse the legitimate security tool INISAFE CrossWeb EX to run shellcode in memory through the Windows binary fsquirt.exe. BleepingComputer (source)
🔴 Exploited or active

CISA/FBI joint advisory on SafePay ransomware

SafePayAA26-293A600+ victims

WhatA joint CISA, FBI and European advisory attributes over 600 critical-infrastructure victims to SafePay ransomware.

DoReview joint advisory AA26-293A.

Details and sources
A joint advisory (AA26-293A) from CISA, the FBI, the U.S. Secret Service, Germany's LKA Rheinland-Pfalz and the UK's NCA/ROCU details SafePay ransomware activity dating back to November 2024, attributing over 600 critical-infrastructure victims to the group, which has filled part of the gap left by BlackCat/LockBit's decline. BleepingComputer (source)

Threat Actor Activity 3

🔴 Exploited or active

China-aligned Storm-0940 abuses agentic-AI pentest tool in ConnectWise supply-chain attack

Storm-0940ARTEXConnectWise Cloud Backup

WhatStorm-0940 used compromised ConnectWise Cloud Backup identities and the ARTEX AI framework to spray 2,250+ tenants and reach privileged cloud access within hours.

DoHunt for high-rate sign-in attempts, OAuth-app abuse and Microsoft Graph credential or privilege changes.

Details and sources
Microsoft attributes a ConnectWise Cloud Backup supply-chain compromise to Storm-0940, a China state-aligned actor active since at least 2021. ConnectWise disclosed unauthorized access to Cloud Backup application identities on October 16 (initial access vector still unexplained); attackers then used ARTEX, a Go-based agentic AI penetration-testing framework, to automate reconnaissance, password spraying and OAuth-app abuse "at machine speed" — a October 10–14 campaign attempted sign-ins against 2,250+ tenants at up to 2,900 attempts/minute, and in multiple instances attackers reached privileged access to a customer's cloud environment in under three hours, with follow-on activity through Microsoft Graph (credential and privilege changes). Hundreds of organizations across aerospace, defense, manufacturing, chemicals, energy, financial services and other sectors have been targeted since September. ConnectWise has suspended Cloud Backup and taken its SaaS Security and Cloud Management services offline; it says existing backup archives remain encrypted and isolated with no sign of compromise. The Hacker News (source) · Dark Reading (source)
🔴 Exploited or active

Flax Typhoon-linked activity via XWiki exploitation

Flax TyphoonXWikiRedTail

WhatVulnCheck tied exploit attempts on five known flaws, including active XWiki exploitation deploying RedTail botnet malware, to Flax Typhoon.

DoPatch internet-facing XWiki instances.

Details and sources
VulnCheck linked exploit attempts against five known vulnerabilities — including active XWiki exploitation deploying RedTail-family botnet malware — to Flax Typhoon, based on source IPs traced to Hong Kong and Taiwan. The Hacker News (source)
🟡 Worth knowing

INTERPOL's "Operation Serengeti III" disrupts cybercrime across Africa

INTERPOLSerengeti III1,462 arrests

WhatAn operation across 21 countries disrupted 10,909 malicious networks and led to 1,462 arrests tied to over $385 million in losses.

Details and sources
A June 29–August 31 2026 law-enforcement operation across 21 countries (mostly in Africa, plus the US) disrupted 10,909 malicious networks, led to 1,462 arrests, identified 65,929 victims, and tied to over $385 million in losses, of which more than $105 million was recovered. The Hacker News (source) · BleepingComputer (source)
‹ 22/10/202626/10/2026 ›