cybr.cx Daily Digest — October 28, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-109618 & CVE-2026-109619 | PHP | CVSS: N/A (KEV listed)
Two PHP vulnerabilities are under active exploitation right now, with CISA's remediation deadline set for tomorrow, October 29. CVE-2026-109618 is an out-of-memory corruption flaw enabling unauthenticated remote code execution; CVE-2026-109619 abuses stale internal request-state to achieve the same outcome. PHP underpins an enormous share of the web — if you're running any PHP-based stack and haven't patched, this is your only priority today.

⚠️ Actively exploited — CVE-2026-48356 | Adobe Commerce & Magento | CVSS: N/A (KEV listed)
An unrestricted file upload vulnerability in Adobe Commerce and Magento is being exploited in the wild, allowing arbitrary code execution in the context of the current user with potential for privilege escalation. E-commerce operators should treat this as critical regardless of CVSS scoring — Magento environments are perennially high-value targets for skimmer and payment-data theft campaigns. Patch or isolate immediately; CISA's due date is November 10.

⚠️ Actively exploited — CVE-2026-111497 | Google Brotli | CVSS: N/A (KEV listed)
Google's Brotli compression library has an unchecked return value flaw that lets a remote, unauthenticated attacker corrupt adjacent decoder state memory via a crafted compressed stream — leading to memory disclosure, denial of service, or potential RCE. Brotli is deeply embedded across web servers, CDNs, browsers, and application runtimes, which makes the blast radius here unusually wide. Audit your dependency trees and update Brotli across all surfaces.

⚠️ Actively exploited — CVE-2026-114625 | Arista VeloCloud Orchestrator | CVSS: N/A (KEV listed)
A SQL injection vulnerability in Arista's on-prem VeloCloud Orchestrator allows remote attackers to access privileged internal functionality and fully compromise the orchestrator host. This is SD-WAN infrastructure — successful exploitation can affect the confidentiality, integrity, and availability of every site the orchestrator manages. CISA's due date has already passed (October 26); if you haven't patched, you're overdue.

⚠️ Actively exploited — CVE-2026-96652 / 96654 / 96655 / 96656 | Plex Media Server | CVSS: N/A (KEV listed)
Four chained Plex vulnerabilities are being actively exploited together: two SSRF flaws (including one requiring only any valid X-Plex-Token), a URI scheme injection allowing cross-plugin function calls, and an arbitrary file write that executes on load. The chain represents a credible full-compromise path from an authenticated (or lightly authenticated) position. Plex is widely self-hosted on home and small-business networks that often share segment space with sensitive systems — patch now.

⚠️ Actively exploited — CVE-2026-110447 | Collabora Online (CODE Server) / Nextcloud | CVSS: N/A (KEV listed)
When the built-in CODE Server component isn't running within a Nextcloud deployment, an OS command injection vulnerability allows unauthenticated remote code execution. Nextcloud installations are common in enterprise and government environments specifically because they're considered a private-cloud alternative — the irony of a public RCE in that context is not lost. Update Collabora Online immediately and verify the CODE server is running and healthy.

⚠️ Actively exploited — CVE-2026-106931 | EG Secure Solutions SiteGuard Server Edition | CVSS: N/A (KEV listed)
SiteGuard Server Edition — a web application firewall product — contains an OS command injection flaw that can hand an attacker root or SYSTEM privileges. Having your security appliance become the attack vector is a particularly bad day. CISA's due date has passed; prioritise patching or replacement.

⚠️ Actively exploited — CVE-2026-110936 | Siemens Software Center | CVSS: N/A (KEV listed)
An uncontrolled search path element flaw in Siemens Software Center allows an authenticated local attacker to drop a malicious DLL into a searched directory and execute arbitrary code at the application's privilege level. Relevant to any OT/ICS environment running Siemens tooling — lateral movement from an IT foothold into Siemens software infrastructure is a realistic scenario here.

CVE-2026-116748 | Transmission BitTorrent Client (< 4.1.3) | CVSS 8.8
A permissive CORS misconfiguration in Transmission's RPC server exposes the X-Transmission-Session-Id anti-CSRF token to any web origin via the 409 handshake response. Any malicious web page can read the token and replay it to issue arbitrary RPC calls — triggering post-download script execution, changing download directories, or deleting torrents and data. Update to 4.1.3 and restrict RPC access to localhost if you haven't already.

CVE-2026-75891 | Kirki Plugin for WordPress (≤ 6.2.5) | CVSS 8.8
A missing capability check on the kirki_post_apis() AJAX dispatcher allows any subscriber-level authenticated user to trigger PHP object injection via deserialization, leading to privilege escalation. With Kirki being a widely used theme customisation framework, the subscriber-to-admin path here is particularly dangerous on multi-user WordPress installations. Update immediately and audit for unexpected admin account creation.

CVE-2026-112351 | Sunhillo SureLine (< 9.1.5) | CVSS 8.8
A bypass of the previously issued fix for CVE-2021-36380: attackers can use URL-encoded newlines and other metacharacters to inject OS commands via the ipAddr or dnsAddr parameters in networkDiag.cgi, executing as root with any valid web account. Incomplete sanitisation patches are a recurring theme in embedded/appliance software — if you're running SureLine, treat the prior fix as insufficient and update to 9.1.5.

Headline News

PHP Under Active Fire: Exploitation in the Wild Accelerates

Telemetry from honeypots and production environments alike is showing sustained exploitation attempts against the two actively exploited PHP RCE vulnerabilities now on CISA's KEV list. Analysis of exploitation patterns shows attackers targeting PHP-FPM and CGI-mode deployments in particular, with payloads ranging from web shells to cryptominer droppers. The breadth of affected infrastructure is staggering — PHP powers an estimated 75% of server-side web applications, meaning the attack surface dwarfs most platform-specific vulnerabilities. Practitioners should not wait for vendor guidance cycles here: patch, validate your PHP version strings externally, and review WAF rules to detect in-progress exploitation. CISA's October 29 due date for federal agencies reflects the urgency, but the risk is equally acute for commercial and self-hosted environments.

LiteLLM Supply Chain Attack: One Forgotten GitHub Key, Many Downstream Victims

A supply chain attack against LiteLLM — a popular open-source proxy and SDK for interacting with large language model APIs — has been traced back to a GitHub master key that was exposed but never rotated. The unrevoked credential gave attackers persistent access to the project's repository, enabling them to introduce malicious code that propagated to downstream consumers of the package. This incident is a textbook illustration of why secrets rotation and exposure monitoring are non-negotiable, not optional hygiene: the key was presumably flagged or visible at some point, yet the remediation step of revocation was never taken. For practitioners, the LiteLLM compromise is also notable because of who uses it — AI-adjacent tooling is now firmly embedded in development and production pipelines, meaning the supply chain risk surface for ML infrastructure deserves the same scrutiny as any other critical dependency. Audit your AI/ML package dependencies for unexpected updates and review your GitHub secret scanning and rotation policies.

South Korea's First AI-Assisted Hacking Case: No IT Background Required

South Korean police have arrested seven individuals in what authorities are describing as the country's first significant case of AI-assisted cybercrime, centred on a 32-year-old with no formal IT background who allegedly used AI tools to compromise approximately 90 hospitals, schools, and companies, extorting roughly $270,000 in total. The case is significant not for its scale but for what it signals about the democratisation of attack capability: the technical barrier to conducting a sustained, multi-target intrusion campaign appears to have dropped to near zero when AI tooling handles reconnaissance, exploitation guidance, and payload generation. Healthcare and education institutions — already chronically under-resourced on defence — were the primary victims, which tracks with broader targeting trends. Defenders should treat "low sophistication attacker" as a meaningfully outdated risk category; the tooling gap between offence and defence is narrowing in the wrong direction.

Schrödinger's Feed

The GQI Q3 2026 State of Play report is flagging something practitioners should take seriously: industry analysts are revising Q-Day timelines downward, and governments are reportedly accelerating post-quantum cryptography migration roadmaps in response. The report notes a market-wide shift toward second-generation, fault-tolerant quantum architectures, which represents genuine engineering progress rather than just investment noise. IBM's concurrent announcement of meter-scale superconducting L-couplers achieving 99% state transfer fidelity between two 120-qubit processors is exactly the kind of hardware milestone that makes those revised timelines credible. If your organisation's PQC transition plan assumes Q-Day is comfortably a decade away, this quarter's data suggests that assumption deserves a revisit.

/dev/random

Tailscale has published a post-incident investigation into a bug in Tailscale SSH — the feature that lets you SSH to nodes on your tailnet without managing keys — where, under specific conditions, the wrong node's SSH session could be presented to the wrong user. In other words: the mesh VPN designed to make access control simpler managed to briefly confuse whose shell is whose, which is roughly the one thing you'd most like it to get right. Tailscale has been commendably transparent about the root cause and remediation steps, which is more than can be said for many vendors facing an identity-confusion bug in their flagship access feature. The lesson, as always: zero-trust is a philosophy, not a guarantee — read your vendor's incident reports.