Koffei

Daily CyberNews · 26/10/2026

‹ 23/10/202627/10/2026 ›
Today in 30 seconds
  • 🔴 Update Collabora Online to 26.4.303: CVE-2026-110447 web shells hit Nextcloud servers.
  • 🔴 Patch Arista VeloCloud Orchestrator: CVE-2026-114625 is exploited and in KEV.
  • 🔴 Patch PHP: PHP2Shell breached a European government; set session.upload_progress.enabled=0 for CVE-2026-115560.
  • 🔴 Block device-code flow via Conditional Access: M365 token replay campaign wipes MFA methods.
  • 🔴 Audit GitHub for ci-sync- branches: Gheist steals CI/CD secrets with stolen tokens.

Vulnerabilities 6

🔴 Exploited or active

Collabora Online web shells on Nextcloud

CVE-2026-110447CVSS 9.2in KEVNextcloud

WhatA command injection in Collabora Online's built-in CODE server via proxy.php is used to plant web shells that search for credentials and crypto wallets.

DoUpdate Collabora Online to 26.4.303 and check Nextcloud servers for web shells.

Details and sources
CISA added CVE-2026-110447 (OS command injection in Collabora Online's built-in CODE server, used with Nextcloud, via proxy.php when the local document server isn't running; CVSS 4.0 9.2, fixed in version 26.4.303 since September) to its Known Exploited Vulnerabilities catalog on Oct 23 with an Oct 26 federal remediation deadline. Attackers are planting web shells on unpatched Nextcloud servers to search for credential stores, password-manager data, and crypto wallets. The Hacker News (source)
🔴 Exploited or active

Arista VeloCloud Orchestrator SQL injection, third flaw exploited since July

CVE-2026-114625CVSS 9.9in KEVVeloCloud

WhatAn authenticated SQL injection in Arista VeloCloud Orchestrator is exploited, the third VeloCloud flaw exploited since July after CVE-2026-16812 and CVE-2026-93952.

DoUpgrade to 5.2.3.18, 6.1.3.10 or 6.4.2.12 and look for connections from 74.82.197[.]86 and 38.54.50[.]151.

Details and sources
CVE-2026-114625 (SQL injection reachable with an authenticated low-privilege session; CVSS v3.1 9.9 / v4.0 9.4) is also now in CISA's KEV catalog (added Oct 23, due Oct 26). Arista confirmed active exploitation and fixed it in 5.2.3.18/6.1.3.10/6.4.2.12; two Japan-geolocated IPs (74.82.197[.]86, 38.54.50[.]151) were shared as IOCs. It's the third VeloCloud Orchestrator bug exploited in the wild since July, after CVE-2026-16812 and CVE-2026-93952. The Hacker News (source) · BleepingComputer (source)
🟠 High impact

Brotli decoder memory-corruption bug hits Node.js, Bun, Envoy, Istio, Python

CVE-2026-111497CVSS 9.2Brotli

WhatAn out-of-bounds write in Google's Brotli library is reachable wherever decompression runs automatically, including Node fetch, Envoy, Istio and Python; no exploitation or reliable PoC.

DoUpdate Brotli to 1.0.10, 1.1.1 or 1.2.1, Node to 22.23.4, 24.22.1 or 26.12.1, Bun to 1.4.4, and Envoy/Istio.

Details and sources
CVE-2026-111497 is an out-of-bounds write in Google's Brotli decompression library (CVSS 4.0 9.2; Red Hat rates it 8.1) triggered by crafted compressed data, reachable anywhere Brotli decompression runs automatically — Node fetch(), Express 5/body-parser 2, Envoy/Istio, or simply having Python Brotli/brotlicffi installed. Fixed upstream (Brotli 1.0.10/1.1.1/1.2.1; Node 22.23.4/24.22.1/26.12.1; Bun 1.4.4; Envoy/Istio point releases). No in-the-wild exploitation; Imperva's red team got code execution in a lab-only build ("Node2Shell") but hasn't released it, and no reliable public PoC exists yet. The Hacker News (source)
🔴 Exploited or active

PHP2Shell (CVE-2026-109618 / CVE-2026-109619) update

PHP2ShellCVE-2026-109618CVE-2026-115560unpatched

WhatA disputed public PoC circulates with widespread exploit attempts, and PHP disclosed a new unpatched session-handler RCE, CVE-2026-115560.

DoPatch PHP and set session.upload_progress.enabled=0 until a fix for CVE-2026-115560 ships.

Details and sources
a public proof-of-concept is now circulating, though its reliability is disputed (WatchTowr says it reproduced CVE-2026-109618 but is withholding technical details; a claimed working PoC posted on X may only work in lab conditions). Multiple firms — including Amazon, which reportedly saw attempts four days before patches shipped — have observed exploitation attempts; Sophos's own telemetry shows scanning and a handful of failed attempts with "no evidence of a successful compromise" through that specific telemetry. Separately, PHP disclosed a new, distinct, still-unpatched RCE risk, CVE-2026-115560 (Oct 25): it requires an app that sets session.serialize_handler to php_binary/php_serialize and then switches it to php at runtime with ini_set() before session_start(), with session upload progress left enabled. PHP's guidance is to set session.upload_progress.enabled=0 (the same mitigation already recommended for CVE-2026-109619) until a routine bugfix release ships a real fix. SecurityWeek (source) · Sophos (source) · The Hacker News (source)
🟡 Worth knowing

Unpatched Tailscale SSH privilege-escalation research ("Tailworm")

Tailscale SSHTailwormno CVE

WhatA researcher showed three lab techniques for privilege escalation and rogue enrollment against Tailscale SSH, all needing an attacker-controlled device.

DoEnable Tailnet Lock, which blocks the rogue device enrollment technique.

Details and sources
independent researcher Erica Windisch (Cyberdione Labs) published three lab-demonstrated techniques against Tailscale 1.104.2/headscale 0.29.5 — root-socket abuse via SSH port forwarding, socket replacement for direct root execution, and a TOCTOU race to capture an enrollment key for rogue device enrollment (blocked when Tailnet Lock is enabled). All require a device the attacker already controls. No CVE yet, no public PoC; Tailscale says it's investigating fixes but isn't recommending anyone disable Tailscale SSH in the meantime. The Hacker News (source)
🟠 High impact

Critical Wasmtime sandbox-escape pair

Added after publishingCVE-2026-115889CVE-2026-115893CVSS 9.9Wasmtime

WhatTwo Cranelift compiler flaws in Wasmtime could let a malicious WebAssembly module escape its sandbox; no exploitation reported.

DoUpdate Wasmtime to 36.0.19, 48.0.7, 49.0.4 or 50.0.0 and rebuild anything embedding it as a library.

Details and sources
the Bytecode Alliance disclosed two critical (CVSS 9.9) flaws in Wasmtime's Cranelift compiler that could let a malicious WebAssembly module escape its sandbox. CVE-2026-115889 (all architectures) is a 24-bit value-tracking overflow in Cranelift before v43 that can cause type confusion; CVE-2026-115893 (ARM64 only, dating back to 2020) lets an out-of-range br_table index jump to arbitrary code when a table has 256,000+ targets. No in-the-wild exploitation and neither is in CISA KEV as of today. Fixed in 36.0.19, 48.0.7, 49.0.4, and 50.0.0 (Oct 19–20); unsupported lines (e.g. 24.x) got no fix, and anything embedding Wasmtime as a library must be rebuilt, not just have its CLI updated. The Hacker News (source)

Breaches / Incidents 3

🟡 Worth knowing

Albert Heijn / Ahold Delhaize (Netherlands) cyberattack — cost estimate revised sharply upward

Ahold DelhaizeAlbert Heijn€120 to 150M

WhatExperts now estimate Albert Heijn's revenue loss from the Oct 11 incident at about €120 to 150 million; home delivery is still down.

Details and sources
two Dutch retail-economics experts (Laurens Sloot, U. Groningen; Erik Hemmes), speaking to RTL Z, now estimate Albert Heijn's revenue loss from the Oct 11 incident at roughly €120–150 million, plus about €5 million in lost pre-tax profit and €1 million in extra costs — well above the earlier "exceeds €10 million" figure; Ahold Delhaize itself hasn't commented on the financial impact. As of the latest update, grocery home delivery was still not possible, though Gall & Gall and Etos sites are back and Belgian webshops are fully operational; Ahold Delhaize still describes it only as caused by "malicious actions of others," with no attribution and no confirmed evidence of customer data theft. NL Times (source) · IBTimes UK (source)
🔴 Exploited or active

PHP2Shell exploitation escalates to a confirmed government breach

PHP2ShellCVE-2026-109618CVE-2026-109619government breach

WhatUnit 42 reports an attacker breached a European government entity via PHP2Shell, taking password hashes and MFA secrets and attempting Domain Admin escalation.

DoPatch PHP and hunt for NTLM-relay tunnels into Active Directory from PHP servers.

Details and sources
Unit 42 (Palo Alto Networks) reported that an unidentified, possibly agentic-AI-driven attacker successfully breached a European government entity via the CVE-2026-109618/CVE-2026-109619 ("PHP2Shell") flaws, obtaining password hashes and MFA secrets within two hours, setting up an NTLM-relay tunnel into Active Directory, and attempting Domain Admin escalation — a step up from "attempts only." Unit 42 did not attribute the intrusion to a specific actor; see Threat Actor Activity below for CYFIRMA's separate, suspected-North-Korea scanning activity against the same flaw. The Hacker News (source)
🟠 High impact

Quest Diagnostics confirms patient records and PII exposed

Added after publishingQuest DiagnosticsQuanum LSMpatient data

WhatAn unauthorized party accessed Quest's Quanum Lab Services Manager platform, exposing health information and PII; victim count is not published.

Details and sources
Quest confirmed that an unauthorized party accessed its Quanum Lab Services Manager (LSM) platform (used by providers to order tests and view results); Quest learned of the access Oct 21 and has now confirmed health information and personally identifiable information were affected, though it hasn't published a victim count or a full list of compromised data types. The platform was paused, core business operations continue, and Quest plans to notify patients, providers, and health plans, offering credit/identity monitoring once the investigation wraps up. No attacker named, no attack method disclosed. Malwarebytes (source)

Malware / Ransomware 6

🔴 Exploited or active

LiteLLM supply-chain compromise — root cause traced to an unrevoked master key

LiteLLM1.107.0leaked master key

WhatA LiteLLM master key leaked in May was used to obtain a GitHub OAuth token and push the malicious 1.107.0, whose backdoor lives outside the app directory.

DoOn hosts that ran LiteLLM 1.107.0, remove the separate Unix backdoor too and rotate SSH, Git, cloud and AI API credentials.

Details and sources
new reporting details how an attacker used a LiteLLM master key that had been committed to a public GitHub repo back in May and never revoked, pulling a GitHub OAuth token on Oct 10 and using it on Oct 21 to push a malicious LiteLLM 1.107.0 release (PyPI, plus matching Docker Hub/GHCR images). The payload runs at Python startup, steals SSH/Git credentials, environment variables, and API keys for several AI and cloud services, and drops a Unix backdoor that polls hourly for attacker code — Microsoft noted the backdoor lives outside the app directory, so removing just the app doesn't remove it. PyPI quarantined the release. The Hacker News (source)
🟡 Worth knowing

AI coding agents hijacked via a planted AGENTS.md/CLAUDE.md file

AGENTS.mdCLAUDE.mdAI coding agentsPoC

WhatResearchers showed a planted AGENTS.md or CLAUDE.md in Downloads can redirect AI coding agents run there to install malware or leak OAuth tokens.

DoAvoid running AI coding agents from the Downloads folder or other untrusted directories.

Details and sources
researchers at Forever Security showed that a malicious instructions file dropped into a victim's Downloads folder can silently redirect an AI coding agent run from that folder — OpenAI Codex followed the planted instructions to install a malicious npm package, function-color, in over 90% of tests, while a malicious CLAUDE.md got Claude Code to redirect its own API traffic to an attacker-controlled endpoint and expose its OAuth token roughly half the time. This is a proof-of-concept, not a confirmed in-the-wild attack. BleepingComputer (source) · The Hacker News (source)
🟡 Worth knowing

Abandoned Popcorn Time updater — a latent multi-million-device botnet risk

Time4PopcornCVE-2026-30612CVSS 9.8

WhatPopcorn Time's updater ran unsigned updates over an unencrypted channel with SYSTEM/root rights from lapsed domains, which Eye Security registered and neutralized.

DoRemove Time4Popcorn from any machine still running it.

Details and sources
Eye Security disclosed ("Operation IXIM," at Hack.lu) that Time4Popcorn's background updater fetched and ran updates over an unencrypted channel with SYSTEM/root privileges without verifying the sender — tracked as CVE-2026-30612 (CVSS 9.8) — and that its lapsed update-server domains could have been re-registered by anyone to push code to every machine still checking in (telemetry suggested 2.5–3M+ Windows PCs and 250,000–350,000 Macs). Eye Security registered the domains itself and pushed over 2 million benign "kill" updates between November 2025 and March 2026 before handing the domains to the Shadowserver Foundation. Not CISA KEV-listed; no evidence of malicious exploitation before Eye Security's intervention. The Hacker News (source)
🟠 High impact

uCreator — rentable Android spyware targeting Telegram and WeChat

Added after publishinguCreatorAndroid spywareTelegramWeChat

WhatuCreator is rentable Android spyware built from tampered Telegram and WeChat APKs that steals 2FA passwords and chats and hijacks accounts.

DoInstall Telegram and WeChat only from official app stores.

Details and sources
researcher Ovi (with back-end code found by Jurre) detailed uCreator, a spyware-for-rent kit built from tampered Telegram/WeChat APKs distributed via sideloading, fake app-store sites, third-party stores (APKPure, ApkSum, AndroidP), and the kit's own "Onion Box" container app. It can steal Telegram 2FA passwords and chat history, reset sessions, hijack the linked phone number, and clone WeChat accounts via stolen login data; the kit has code targeting roughly 20 apps. Ovi links parts of the operation (unproven) to three companies in Hunan province, China, and assesses it as financially motivated; targeting is unconfirmed but likely mobile-wallet and messaging-app users in China and Southeast Asia. The Hacker News (source)
🔴 Exploited or active

"Gheist" automated campaign abuses stolen GitHub tokens to leak CI/CD secrets

GheistGitHub tokensCI/CD secrets

WhatAn active campaign uses stolen GitHub tokens to plant ci-sync- branches whose workflows print secrets into Actions logs, hitting 850+ accounts.

DoSearch for ci-sync- branches and deleted Actions runs, rotate exposed CI/CD secrets and revoke leaked GitHub tokens.

Details and sources
Wiz Research has tracked Gheist (named for its "gheist" API user agent) since Sept 15 — it uses stolen GitHub tokens (sourced from infostealer logs and prior supply-chain breaches) to plant a ci-sync--prefixed branch with a workflow that prints job secrets into the Actions log (base32-encoded to dodge GitHub's masking), downloads the log, then deletes the run within seconds — exfiltrating entirely through GitHub itself, with no attacker server to block. Some runs also cloned private repos first. Wiz recovered 13,500+ planted branches across 850+ compromised accounts, spanning personal projects to enterprise private repos and GitHub Enterprise Server; the campaign was still active and evolving as of Oct 24. Distinct from the unrelated GhostAction campaign; no actor attribution given, though some reused accounts overlap with activity researchers tie to North Korea. The Hacker News (source)
🔴 Exploited or active

"Token spraying" campaign hijacks Microsoft 365 accounts via replayed AitM tokens

Microsoft 365AitM token replay193.26.115.0/24

WhatAttackers replay M365 tokens from AitM phishing and device-code lures, then wipe MFA methods, register rogue Entra devices and hide payroll emails.

DoRestrict device-code flow with Conditional Access, require MFA unconditionally and review new authenticator apps, Entra devices and inbox rules.

Details and sources
Huntress documented a campaign (observed July 14–Sept 22) where attackers harvest Microsoft 365 access tokens via adversary-in-the-middle phishing pages and device-code social engineering, then replay the tokens from their own devices to sign in without needing MFA again. Once in, they registered their own authenticator app, wiped victims' existing MFA methods within three hours at 19 organizations, registered at least 48 rogue Entra devices for durable access, and set up hidden inbox rules to hide payroll/deposit/bonus emails. Scale: 497 successful sign-ins out of 200,000+ attempts, 97 compromised accounts at 79 organizations. Attacker infrastructure traces to the 193.26.115.0/24 range, previously linked to StarkRDP/rdp.sh. No Microsoft-side fix; Huntress recommends Conditional Access restrictions on the device-code flow and unconditional MFA. The Hacker News (source)

Threat Actor Activity 2

🔴 Exploited or active

APT41-linked "BlackX" UEFI bootkit bypasses Secure Boot via an expired-but-unrevoked certificate

APT41BlackXCVE-2022-21894UEFI

WhatAPT41's BlackX bootkit uses Baton Drop and the expired but unrevoked Windows Production PCA 2011 certificate to bypass Secure Boot on patched systems.

DoApply the Secure Boot DBX revocation step for vulnerable boot binaries, not just the regular patch.

Details and sources
Kaspersky attributes (medium confidence) a new UEFI bootkit, BlackX, to APT41, used since early 2026 against diplomatic targets in Tajikistan, Uzbekistan, and Turkmenistan. It exploits the old CVE-2022-21894 "Baton Drop" Secure Boot bypass together with the fact that the "Windows Production PCA 2011" certificate wasn't added to the Secure Boot DBX revocation list before it expired Oct 19 — so a vulnerable boot binary still loads on fully patched systems that hadn't manually applied the extra revocation step. It loads a malicious kernel driver (signed by a now-revoked Chinese shell-company certificate), deploys a RAT using Google Drive/OneDrive/Yandex.Disk for C2, and shares a code-overlap indicator with FishMonger/Aquatic Panda, both tracked under the APT41 umbrella. The Hacker News (source) · BleepingComputer (source)
🔴 Exploited or active

Suspected North Korea-linked scanning for PHP2Shell, separate from the confirmed government breach

PHP2ShellLazarus Groupscanning

WhatCYFIRMA links low-volume PHP2Shell scanning across 21 nodes against 216 partially patched hosts to a Lazarus-ecosystem operator, with moderate confidence.

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

Details and sources
CYFIRMA (moderate confidence) linked low-volume scanning — rotated across 21 compromised nodes against 216 partially-patched hosts in government, finance, aviation, and logistics — to a Korean-speaking, Lazarus Group-ecosystem operator, citing reused Lazarus C2 infrastructure and I2P chat language. This is a distinct claim from Unit 42's confirmed European-government breach (see Breaches / Incidents above), which Unit 42 itself did not attribute to North Korea. The Hacker News (source)
‹ 23/10/202627/10/2026 ›