cybr.cx | Daily Digest — October 24, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-114625 | Arista VeloCloud Orchestrator | No CVSS listed (KEV)
A SQL injection flaw in Arista's VeloCloud Orchestrator on-premises deployment is being actively exploited right now. Remote attackers can access privileged internal functionality and fully compromise the orchestrator host — confidentiality, integrity, and availability all at risk. CISA's remediation deadline was yesterday (Oct 23). If you're running VCO on-prem and haven't patched, treat this as a fire drill.

⚠️ Actively exploited — CVE-2026-110447 | Collabora Online (Built-in CODE Server) | No CVSS listed (KEV)
An OS command injection vulnerability in Collabora Online's built-in CODE Server allows remote code execution when the CODE server process isn't running — a condition that can arise during restarts or crashes. This is particularly dangerous in Nextcloud deployments, where Collabora is a common self-hosted productivity integration. Patch or disable immediately; CISA's due date was also Oct 23.

⚠️ Actively exploited — CVE-2026-96652 / 96654 / 96655 / 96656 | Plex Media Server | No CVSS listed (KEV)
Four chained vulnerabilities in Plex Media Server are being exploited in combination. The chain includes two SSRF flaws (allowing arbitrary internal/external POST and GET requests), a URI scheme injection enabling cross-plugin function calls, and an arbitrary file write reachable by admin users that executes on load. Any valid X-Plex-Token can trigger the SSRF entry point — raising the effective attack surface considerably. Patch Plex Media Server urgently, especially on network-exposed instances.

⚠️ Actively exploited — CVE-2026-106931 | EG Secure Solutions SiteGuard Server Edition | No CVSS listed (KEV)
An OS command injection flaw in SiteGuard Server Edition allows attackers to run arbitrary commands with root or SYSTEM privileges. The due date (Oct 23) has already passed. If SiteGuard is in your stack, this is priority one.

⚠️ Actively exploited — CVE-2026-110936 | Siemens Software Center | No CVSS listed (KEV)
A DLL search-order hijacking vulnerability allows an authenticated local attacker to drop a crafted DLL into a directory searched by the application at runtime, achieving code execution with the application's privilege level. Exploitation is confirmed in the wild; due date is Nov 3, but don't wait.

⚠️ Actively exploited — CVE-2026-102764 | BeyondTrust Remote Support | No CVSS listed (KEV)
An authentication bypass-by-spoofing vulnerability in BeyondTrust Remote Support lets remote attackers perform file operations on the appliance without valid credentials. Given that Remote Support tools are high-value targets for initial access, treat this as critical regardless of its listed severity.

⚠️ Actively exploited — CVE-2025-61624 | Fortinet FortiOS / FortiPAM / FortiProxy / FortiSwitchManager | No CVSS listed (KEV)
A path traversal flaw across several Fortinet products lets authenticated admins (with read-write permissions) write or delete arbitrary files via CLI commands. In environments with delegated admin access, this significantly expands the blast radius of a compromised account.

CVE-2026-57845 | Hugging Face Datasets ≤ 5.1.0 | CVSS 8.8 — HIGH
An insecure deserialization vulnerability in the deprecated pandas packaged builder allows an attacker to craft a malicious pickle file that executes arbitrary Python code when a victim loads the dataset. No authentication required on the attacker's side — just a poisoned dataset. ML pipelines ingesting third-party datasets from Hugging Face Hub are directly exposed. Avoid loading untrusted datasets with affected versions and upgrade immediately.

CVE-2026-115726 | Aider ≤ 0.86.2 | CVSS 7.8 — HIGH
Aider, the popular AI coding assistant, auto-loads .aider.conf.yml from the working directory without a trust prompt. A malicious repository can set notifications-command or editor values that are passed to subprocess with shell=True, executing arbitrary commands with the developer's privileges. Supply-chain and repo-cloning scenarios are obvious attack vectors. Upgrade to a patched release or audit any .aider.conf.yml files before running Aider in unfamiliar repositories.

CVE-2026-108456 | Azure AI Foundry | CVSS 7.7 — HIGH
An SSRF vulnerability in Azure AI Foundry allows an authorised attacker to pivot network requests through the service to reach internal endpoints, with potential for privilege escalation. Cloud-native AI development environments are increasingly attractive targets for lateral movement. Check Azure's advisory for affected configurations and apply available mitigations.

CVE-2026-115484 / 115486 | open-iscsi (iscsiuio) | CVSS 7.6 / 7.1 — HIGH
Two flaws in open-iscsi's iscsiuio component: a DHCPv4 option parser heap over-read (7.6) that enables crash or memory corruption via a crafted DHCP response, and an IPv6 Neighbor Discovery handler (7.1) that allows ARP/NDP table poisoning from the local network segment. Both are exploitable by an attacker with layer-2 adjacency. Storage network segments are frequently under-monitored — patch promptly.

Headline News

PHP Vulnerability Triggers Broad Disruptions in Japan

A critical flaw in PHP — CVE-2026-109618/109619, relating to OOM bailout and stale state corruption — has caused significant real-world disruption across Japan since October 21, affecting banks, restaurant chains, government agencies, and high-profile websites including political figures' public pages. PHP underpins an estimated 70% of web infrastructure globally, making the blast radius of any critical runtime bug enormous. The flaws involve state corruption via bailout conditions that can be triggered under memory pressure, and while official fixes exist for modern PHP branches, many production environments still run PHP 5.x on Enterprise Linux 6 and 7 — versions that required unofficial backports to receive the fix at all. Practitioners should audit their PHP version landscape now: legacy 5.x deployments on EL6/EL7 are disproportionately exposed, and the Japan disruptions demonstrate that theoretical vulnerabilities in runtime internals translate directly to production outages at scale.

China-Linked Storm-0940 Uses AI Pentesting Tool to Compromise Microsoft 365 at Scale

Microsoft has formally attributed a large-scale credential-stuffing and intrusion campaign to Storm-0940, a China state-aligned threat actor, linking them to a breach of ConnectWise infrastructure that provided a launchpad for targeting approximately 2,250 Microsoft 365 tenants. The group deployed an AI-assisted penetration testing tool called ARTEX to accelerate attack cadence — moving from credential acquisition to tenant compromise at machine speed in ways that would be impractical with manual tooling. The use of managed service provider (MSP) infrastructure as a pivot point is a well-established but persistently effective technique, and the integration of AI tooling into offensive operations marks a meaningful shift in threat actor capability. Defenders should review conditional access policies, monitor for anomalous OAuth token activity, and treat any ConnectWise-connected environments as potentially in scope for this campaign.

Dutch Intelligence Had Covert Access to Jared Kushner's Phone

Reporting has confirmed that Dutch intelligence services covertly accessed Jared Kushner's phone, with the compromise revealing details of private negotiations and financial transactions running into the millions. The incident underscores the persistent reality that personal devices used for sensitive communications — even by individuals at the highest levels of government — represent a critical and frequently under-hardened attack surface. From a practitioner standpoint, the case is a reminder that BYOD policies, consumer-grade messaging applications, and informal communication channels remain exploitable regardless of the user's institutional security resources. Nation-state mobile compromises of this kind typically involve either sophisticated spyware, baseband-level exploits, or exploitation of unpatched OS vulnerabilities — the technical specifics here remain undisclosed, but the operational security failures are instructive.

Schrödinger's Feed

Researchers have reached what's being described as the "ultimate quantum limit" in understanding how atomic-scale defects in materials like diamond and silicon carbide behave under optical excitation — uncovering a previously unknown energy-loss mechanism in silicon carbide through a new computational method. This matters because silicon carbide defects are promising candidates for quantum memory and quantum networking nodes, and correctly modelling their decoherence is essential for building reliable quantum repeaters. Unexpected loss pathways in qubit substrate materials are exactly the kind of detail that determines whether post-quantum cryptographic infrastructure built on quantum key distribution holds up in practice. Practitioners building long-term cryptographic roadmaps should note that quantum hardware still surprises researchers at the physics level — the timeline to cryptographically relevant quantum computers remains genuinely uncertain, but the science is advancing faster than the models predicted.

/dev/random

A team of researchers set out with a stated goal — hack the iPhone — and, by their own account, did exactly that. The writeup details a successful exploitation chain targeting Apple's mobile platform, walking through the specific vulnerability classes and bypass techniques required to achieve meaningful code execution. The matter-of-fact title ("Today, we hacked what we'd set out to hack") has the energy of someone checking a very expensive item off a to-do list. The security community's response was immediate and enthusiastic, which suggests that either people appreciate the straightforward honesty, or everyone just enjoys watching someone poke a $1,000 glass rectangle until it does something it shouldn't.