Koffei

Daily CyberNews · 22/10/2026

‹ 21/10/202623/10/2026 ›
Today in 30 seconds
  • 🔴 Rotate UpdraftCentral connection keys: stolen keys are used to backdoor WordPress sites.
  • 🔴 Check for LiteLLM v1.107.0: the release carries a credential stealer and backdoor.
  • 🔴 Patch Plex, SiteGuard and Siemens Software Center: three exploited flaws are in KEV.
  • 🟠 Install PHP 8.2.35, 8.3.36, 8.4.27 or 8.5.12 for CVE-2026-109618 and CVE-2026-109619.
  • 🟠 Patch self-managed GitLab for CVE-2026-110264, CVSS 10.0 RCE with no workaround.

Vulnerabilities 4

🟠 High impact

PHP ships emergency fix for two critical pre-auth RCE flaws

CVE-2026-109618CVE-2026-109619CVSS 9.2

WhatTwo pre-auth PHP memory-corruption flaws allow RCE against Apache mod_php and PHP-FPM; PoC material reportedly circulates but no exploitation is confirmed.

DoUpgrade to PHP 8.2.35, 8.3.36, 8.4.27 or 8.5.12, or the limited backports for 7.4, 8.0 and 8.1.

Details and sources
CVE-2026-109618 (out-of-memory handling memory corruption) and CVE-2026-109619 (use-after-free in retained request state), both CVSS v4.0 9.2, were patched a day earlier than planned after advisory details leaked ahead of the coordinated disclosure. Fixed in PHP 8.2.35, 8.3.36, 8.4.27 and 8.5.12 (with limited backports for 7.4/8.0/8.1); researchers reproduced RCE against Apache mod_php and PHP-FPM. No in-the-wild exploitation is confirmed yet, but PoC material reportedly began circulating after release. Rapid7 (source) · Threat Landscape Blog (source)
🟠 High impact

Critical unauthenticated RCE in self-managed GitLab

CVE-2026-110264CVSS 10.0GitLab

WhatAn integer overflow in GitLab's bundled Onigmo regex engine allows unauthenticated RCE on self-managed GitLab CE/EE from version 7.0; no exploitation is reported.

DoUpgrade to GitLab 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10 or 18.11.13; there is no workaround.

Details and sources
CVE-2026-110264 (CVSS 10.0) is an integer overflow in GitLab's bundled Onigmo regex engine, reachable via a crafted regular expression, affecting self-managed GitLab CE/EE from version 7.0 (2014) onward. GitLab reports no sign of exploitation and the flaw isn't yet in CISA's KEV catalog, but there is no workaround only patching (19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10 or 18.11.13, depending on branch). The Hacker News (source)
🔴 Exploited or active

CISA adds three actively-exploited flaws to its KEV catalog

CVE-2026-96656CVE-2026-106931CVE-2026-110936KEV

WhatCISA added exploited flaws in Plex Media Server, SiteGuard Server Edition and Siemens Software Center to KEV, with federal deadlines Oct 23 to Nov 3.

DoPatch Plex Media Server, SiteGuard Server Edition and Siemens Software Center before the KEV deadlines.

Details and sources
the "Plex2Shell" chain in Plex Media Server headlined by CVE-2026-96656 (arbitrary file write to RCE, CVSS v4.0 8.6), already deployed in cryptominer/proxy-malware honeypot attacks plus CVE-2026-106931, a vendor-confirmed-exploited OS command injection in Japan's SiteGuard Server Edition WAF (CVSS v4.0 8.5), and CVE-2026-110936, a DLL-sideloading flaw in Siemens Software Center (CVSS v4.0 8.4) being abused via ClickFix-style social engineering. Federal remediation deadlines run October 23–November 3. CISA (source) · The Hacker News (source) · Security Affairs (source)
🟠 High impact

Oracle's October Critical Patch Update ships 1,295 fixes

Oracle CPUCVE-2026-75803CVSS 9.8

WhatOracle's October CPU ships 1,295 fixes, led by unauthenticated critical flaws in Database Server, APEX (CVE-2026-101521) and Essbase (CVE-2026-91665).

DoApply the October 2026 Oracle CPU, prioritizing Database Server 19.3 to 19.32 and 23.4 to 23.26.3, APEX 24.1 to 26.1.4 and Essbase 21.8.3.0.0.

Details and sources
three unauthenticated, remotely-exploitable critical flaws rated CVSS 9.8 lead the quarterly CPU: CVE-2026-75803 (an OpenSSL/TLS flaw in Oracle Database Server, versions 19.3–19.32 and 23.4–23.26.3), CVE-2026-101521 (Oracle APEX, versions 24.1–26.1.4), and CVE-2026-91665 (Oracle Essbase 21.8.3.0.0). Oracle (source)

Breaches / Incidents 5

🟠 High impact

GMO's Onamae.com Desktop Cloud hacked, ~26,700 VPS shut down

GMOOnamae.com Desktop Cloud26,739 VPS

WhatA third party installed an unauthorized program on GMO's Desktop Cloud customer VPS environments, and all 26,739 affected servers are being shut down.

DoDesktop Cloud customers should rotate credentials.

Details and sources
GMO Internet confirmed on October 20 that a third party installed an unauthorized program on customer VPS environments of its FX auto-trading "Desktop Cloud" service; 26,739 servers were affected and have been shut down in sequence since October 21, suspending the service with no restart date set. Data on affected servers may have been viewed; no misuse is confirmed yet, but customers are told to rotate credentials. Internet Watch (source) · BleepingComputer (source)
🟡 Worth knowing

Zeabur suffers second breach in two months, will shut down entirely

Zeaburfollow-upshutting down

WhatZeabur's Oct 14 to 15 intrusion exposed about 321,810 accounts and 3.2M environment-variable records, and the company will close by end of December 2026.

Details and sources
Developer-hosting platform Zeabur disclosed an October 14–15 intrusion using credentials likely traced back to an August breach that exposed roughly 321,810 accounts and 3.2 million environment-variable records across its shared clusters. The company announced it will cease operations entirely by the end of December 2026. BleepingComputer (source)
🔴 Exploited or active

Stolen UpdraftCentral keys threaten ~400,000 WordPress sites

UpdraftCentralstolen connection keysWordPress

WhatAttackers stole UpdraftCentral, Vault and Clone connection keys from TeamUpdraft and use them to add rogue admins and malicious plugins to WordPress sites.

DoOn WordPress sites with an UpdraftCentral connection key, check for rogue admin accounts and unknown plugins.

Details and sources
TeamUpdraft (maker of UpdraftPlus/UpdraftCentral) detected unauthorized access to its own site over October 17–18, exposing customer records, password hashes, and UpdraftCentral/Vault/Clone connection keys. Attackers began using the stolen keys on October 18 to add rogue admin accounts and install malicious plugins on sites holding a connection key — roughly 10% of UpdraftPlus's install base. BleepingComputer (source) · The Hacker News (source)
🟡 Worth knowing

State Bank of India confirms a data leak, disputes the claimed scale

State Bank of Indiadata leak

WhatSBI confirmed the loss of 3,897 customer records without account numbers or passwords, while a dark-web seller claims about 375,000 records.

Details and sources
SBI told Indian stock exchanges on October 21 that it "experienced an incident involving the loss of certain data" after a CERT-In/threat-intel alert. It has so far confirmed 3,897 customer records (names, contact details, branch/loan metadata no account numbers, passwords or Aadhaar/PAN data) out of a dataset a dark-web seller claims holds roughly 375,000 records. The Hacker News (source)
🟡 Worth knowing

Albert Heijn parent files police report, most services restored

Ahold DelhaizeAlbert Heijnfollow-up

WhatAhold Delhaize filed a police report over the Oct 11 incident; most online services are restored and no customer data theft is reported.

Details and sources
Ahold Delhaize has filed a police report over the October 11 incident that took down Albert Heijn, Etos and Gall & Gall's online operations, with the CEO describing it as "malicious actions by others." The Albert Heijn website/app and the Gall & Gall/Etos sites are back online, Belgian online sales have fully resumed, and home delivery was due to resume October 23; the company still reports no evidence of customer data theft, and the estimated cost to Albert Heijn alone now exceeds €10 million. NL Times (source)

Malware / Ransomware 5

🔴 Exploited or active

Joint advisory: SafePay ransomware has hit 600+ critical-infrastructure orgs

SafePayAA26-293Acritical infrastructure

WhatA joint advisory says SafePay has hit 600+ critical-infrastructure organizations via credential stuffing, brute force and edge-device exploitation, then wipes shadow copies and NAS backups.

DoReview advisory AA26-293A, harden VPN and edge devices against brute force, and protect NAS backups and shadow copies.

Details and sources
CISA, the FBI, the US Secret Service, and German/UK law enforcement jointly warned (advisory AA26-293A) that SafePay a private, non-affiliate double-extortion group active since August 2024 has hit at least 600 critical-infrastructure organizations across the Americas, Europe and Australia, via credential stuffing, brute force, VPN exploitation (including Mullvad-linked IPs) and automated edge-device CVE exploitation, then sabotaging recovery by deleting shadow copies and resetting NAS backups. BleepingComputer (source)
🔴 Exploited or active

OVHcloud CLI backdoored via long-compromised GitHub account

OVHcloud CLIv0.11.0 to 0.15.0GlassWorm

WhatA long-compromised developer account trojanized OVHcloud CLI releases with a loader that fetches a secret-harvesting second stage via a Solana memo dead drop.

DoIf you installed OVHcloud CLI v0.11.0 to 0.15.0 from GitHub, rotate secrets stored on that machine.

Details and sources
A developer's OVHcloud GitHub account, compromised for 6+ months, was used to trojanize OVHcloud CLI releases v0.11.0–0.15.0 (macOS/Linux/Windows) by exploiting a disabled release-immutability setting; a hidden loader fetched a second stage via a Solana-memo "dead drop" to harvest secrets from disk. OpenSourceMalware links the operation to the ongoing GlassWorm campaign. BleepingComputer (source)
🔴 Exploited or active

LiteLLM's PyPI/Docker release compromised via stolen OAuth tokens

LiteLLMv1.107.0PyPI/Docker

WhatStolen maintainer OAuth tokens were used to publish LiteLLM v1.107.0 with a credential stealer and a backdoor that runs at Python startup.

DoRemove LiteLLM v1.107.0 and rotate cloud tokens, private keys and browser credentials on hosts that ran it.

Details and sources
Attackers used stolen GitHub CLI OAuth tokens from three LiteLLM maintainer accounts to publish a malicious v1.107.0 release carrying a credential stealer ("privcollect") and a backdoor that auto-runs at Python interpreter startup, exfiltrating cloud tokens, private keys and browser data. The Hacker News (source)
🔴 Exploited or active

ClickFix hits record scale, leans on fake AI-tool pages

ClickFixInstallFixLLMshareEtherHiding

WhatClickFix makes up 67% of Push Security phishing detections, now cloning AI-tool install pages and abusing shared AI chats to get victims to paste commands.

DoWarn users never to paste commands from AI-tool install pages or shared chats into a terminal or Run dialog.

Details and sources
ClickFix now accounts for 67% of Push Security's phishing detections (up from a 52% average), with Microsoft reporting 23x growth and 1.1 million+ devices affected since December. New sub-techniques clone Claude/ChatGPT/Codex install pages ("InstallFix") or abuse shared AI chats/Artifacts ("LLMshare") to get victims to paste malicious commands, with campaign configs stored in blockchain smart contracts ("EtherHiding") to resist takedown. Google responded the same day by adding Gemini-powered ClickFix warnings to Chrome. BleepingComputer (source)
🟠 High impact

300+ Tor exit relays hijacked to intercept SSH logins

Tor exit relaysSSH interceptionEbury

What377 Tor exit relays were reconfigured Oct 1 to 5 to capture SSH passwords, and at least one operator was then backdoored with Ebury.

DoChange SSH passwords used over Tor in early October and stop reusing SSH passwords.

Details and sources
Between October 1–5, 377 relays run by 59 operators were reconfigured to redirect SSH traffic (peaking at 7.46% of Tor's exit capacity) to interception servers capturing passwords; on at least one major operator's network, attackers then deployed the Ebury OpenSSH backdoor after a reused-password login. The Tor Project flagged and delisted the relays by October 5; the operator(s) remain unidentified. The Hacker News (source)

Threat Actor Activity 6

🔴 Exploited or active

China-aligned Storm-0940 weaponizes agentic AI pentesting tool against ConnectWise

Storm-0940ARTEXConnectWise Cloud Backup

WhatStorm-0940 compromised ConnectWise Cloud Backup to deploy the ARTEX AI pentesting framework for high-speed password spraying against 2,250+ tenants since September.

DoHunt for high-rate password spraying and privilege escalation linked to ConnectWise Cloud Backup.

Details and sources
Microsoft disclosed that Storm-0940 compromised ConnectWise's Cloud Backup application and used it to deploy ARTEX, a Go-based AI pentesting framework, for machine-speed password spraying (2,250+ tenants targeted, up to 2,900 attempts/minute) and privileged-access escalation against hundreds of organizations in aerospace, defense, energy, government and finance since September. The Hacker News (source)
🟡 Worth knowing

Newly named China-aligned cluster "Bronze Sandcastle" detailed

Bronze SandcastleChina-alignedSandboxie sideloading

WhatSophos detailed a new China-aligned cluster abusing Sandboxie DLL sideloading to run an in-memory backdoor against targets in four countries.

Details and sources
Sophos documented a previously undocumented China-aligned threat cluster, active since mid-2025, abusing Sandboxie DLL sideloading to run an in-memory backdoor against targets in the Philippines, Taiwan, Afghanistan and Namibia. The Hacker News (source)
🟡 Worth knowing

Lazarus Group's "FlashHole" campaign hides payloads with post-quantum crypto

Lazarus GroupFlashHoleSouth Korea

WhatLazarus targets South Korean software, construction and telecom-equipment firms, using ML-KEM, GOST and Threefish to hide its C2 channel.

Details and sources
Kaspersky detailed a North Korean Lazarus Group campaign targeting South Korean software, construction and telecom-equipment firms that uses NIST-standardized ML-KEM post-quantum key exchange alongside GOST and Threefish to obscure its C2 channel a defender-evasion move rather than a new offensive capability. BleepingComputer (source)
🟡 Worth knowing

Pegasus spyware hits Serbian university rector days before snap election

NSO GroupPegasusSerbia

WhatPegasus was found on the Belgrade University rector's phone six days before Serbia's snap election; no operator is named.

Details and sources
Citizen Lab and Serbia's SHARE Foundation found NSO Group's Pegasus spyware on the phone of Belgrade University's rector, Vladan Đokić, six days before Serbia's October 25 snap election; no operator has been named. Security Affairs (source)
🟡 Worth knowing

Nigerian national sentenced to 8 years for tax-fraud and BEC schemes

SDNYtax fraudBECsentencing

WhatA Nigerian national was sentenced for spearphishing US tax preparers and running fraud schemes with over $10 million in losses.

Details and sources
Chukwuemeka Victor Amachukwu was sentenced in the Southern District of New York for spearphishing US tax preparers, filing $8.4 million in fraudulent refund claims, and running fraudulent SBA loan and investment-fraud schemes totaling over $10 million in losses. BleepingComputer (source)
⚪ Context

US orders DeepSeek removed from federal civilian systems

DHSDeepSeekfederal ban

WhatDHS ordered federal civilian agencies to remove DeepSeek products and banned new purchases, citing supply-chain risk.

Details and sources
DHS directed federal civilian agencies to remove Chinese AI company DeepSeek's products and banned new purchases, under the Federal Acquisition Supply Chain Security Act, citing supply-chain risk. The Hacker News (source)
‹ 21/10/202623/10/2026 ›