cybr.cx Daily Digest — September 27, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-87902 | WordPress Core | CVSS: N/A (KEV-listed)
WordPress Core contains a remote file inclusion vulnerability allowing unauthenticated attackers to manipulate page-template resolution to load arbitrary local .php files outside the active theme directories, leading to remote code execution. With WordPress powering roughly 40% of the web, the blast radius here is enormous. CISA's remediation deadline was yesterday — if you haven't patched, stop reading and go do that first.

⚠️ Actively exploited — CVE-2026-65660 | Microsoft SharePoint | CVSS: N/A (KEV-listed)
A code injection vulnerability in Microsoft SharePoint allows an authorised network attacker to execute arbitrary code. SharePoint's position as a core enterprise collaboration platform makes this high-value for lateral movement post-compromise. CISA's federal deadline was September 28 — treat that as your deadline too.

⚠️ Actively exploited — CVE-2026-67279 | MikroTik RouterOS | CVSS: N/A (KEV-listed)
An improper workflow enforcement flaw in RouterOS lets unauthenticated clients open session channels and send exec requests — and it chains with CVE-2026-86060 for unauthenticated full exploitation. MikroTik devices are deeply embedded in ISP and enterprise networks globally and have a long history of being absorbed into botnets. Patch or isolate management interfaces immediately.

⚠️ Actively exploited — CVE-2026-5430 | WSO2 API Control Plane, API Manager, Traffic Manager & Universal Gateway | CVSS: 10.0
A path traversal flaw enabling unrestricted file upload and remote code execution across WSO2's API management stack. This one scores a perfect 10, has been exploited for at least 12 days in the wild, and a patch has been available since April. There is no excuse. See the Headline News section for fuller context.

⚠️ Actively exploited — CVE-2026-71362 | Adobe Commerce / Magento | CVSS: N/A (KEV-listed)
An incorrect authorisation vulnerability allowing attackers to gain elevated access to sensitive resources without any user interaction. Magento storefronts are perennial targets for skimming and data theft; elevated access without interaction is as bad as it sounds. CISA's deadline passed September 27.

⚠️ Actively exploited — CVE-2026-94127 | F5 BIG-IP APM | CVSS: N/A (KEV-listed)
A heap-based buffer overflow in BIG-IP APM triggers when both an access policy and an OAuth profile are configured on a virtual server, enabling unauthenticated remote code execution. F5 devices sitting at the network edge with OAuth exposed to the internet are the exact targets threat actors scan for first. Patch or disable the vulnerable configuration combination now.

⚠️ Actively exploited — CVE-2026-85102 & CVE-2026-93616 | Check Point Security Gateway / Management Server | CVSS: N/A (KEV-listed)
Two separate actively exploited Check Point flaws: an improper certificate validation bug in Security Gateway and Spark Firewall (Site-to-Site and Remote Access VPN) enabling unauthenticated RCE, and a path traversal in Security Management Server and related products allowing unauthenticated script upload and execution. Two critical bugs, same vendor, both in the wild — Check Point environments should be treated as priority patches this weekend.

⚠️ Actively exploited — CVE-2026-93952 | Arista VeloCloud Orchestrator | CVSS: N/A (KEV-listed)
Improper input validation in on-prem VCO deployments allows a remote attacker to reach privileged internal functionality, compromising confidentiality, integrity, and availability of the orchestrator. SD-WAN orchestrators have full visibility and control over branch connectivity — a compromised VCO is a compromised network fabric.

⚠️ Actively exploited — CVE-2026-7273 | Zyxel GS1900 Series Switches | CVSS: N/A (KEV-listed)
A stack-based buffer overflow in the CGI program of GS1900 series switches allows a LAN-based unauthenticated attacker to execute OS commands via a crafted HTTP request. LAN-based exploitation might sound limited, but these are access-layer switches — once an attacker has LAN access, this is trivial pivoting. CISA's deadline was September 24.

CVE-2026-100596 / 100587 / 100586 / 100580 / 100575 / 100552 / 100544 | OpenClaw (npm) | CVSS: 8.8 (multiple)
A cluster of eight HIGH-severity vulnerabilities across the OpenClaw AI agent framework and its integrations (voice call, Slack, Codex runtime). The flaws collectively allow non-owner or unauthenticated users to bypass tool policies, install arbitrary plugins, execute shell commands, create persistent cron jobs, and gain access to files and processes with OpenClaw's privileges. If you're running any OpenClaw component in production, upgrade to at least 2026.8.1 immediately — the attack surface spans chat, voice, Slack, and scheduled tasks.

CVE-2026-100520 | Laranode (before 1.2.1) | CVSS: 8.8
A path traversal vulnerability in the file manager upload endpoint allows authenticated users to write PHP files into other tenants' web roots and execute code under their identity — a serious multi-tenant isolation failure. Shared hosting or SaaS deployments of Laranode should upgrade to 1.2.1 without delay.

Headline News

WSO2's Perfect-10 JWT Bypass: A Masterclass in How Not to Handle Patch Urgency
CVE-2026-5430, a JWT authentication bypass affecting WSO2 API Manager, API Control Plane, Traffic Manager, and Universal Gateway, carries a CVSS score of 10.0 — the maximum — and has been actively exploited for at least 12 days as of this writing. The patch shipped in April 2026, meaning exposed organisations had roughly five months to act before exploitation began. A forged or manipulated JWT token can bypass authentication entirely across WSO2's API management stack, giving attackers unauthenticated access to APIs, administrative interfaces, and backend services. CISA's federal remediation deadline expired today. If your organisation runs WSO2 products and hasn't patched, assume you should already be in incident response mode.

ShinyHunters Turns the Tables on Clop — Hacks the Hackers via Unpatched CMS
In a development that would be darkly comedic if the stakes weren't so high, the Clop ransomware gang's own data leak site was compromised and defaced — by ShinyHunters, exploiting an unauthenticated path traversal vulnerability in Grav CMS. Clop has since migrated its Tor infrastructure to a new address, effectively confirming the breach. The irony of a ransomware operation running an unpatched content management system isn't lost on anyone, but the tactical implications matter: threat actor infrastructure is itself attackable, and rival groups are actively probing it. For defenders, this underscores that even adversary-operated systems follow the same patch-or-perish rules — and that the threat landscape includes inter-gang conflict that can disrupt leak timelines and victim communications.

OpenAI Agents Used to Compromise Hugging Face Infrastructure
Detailed post-incident analysis has emerged describing how OpenAI-based agents were used to compromise systems at Hugging Face, the widely-used machine learning platform hosting hundreds of thousands of models and datasets. The attack reportedly leveraged agentic AI capabilities to autonomously probe, exploit, and move laterally within Hugging Face's environment — marking one of the first publicly documented cases of AI agents being used end-to-end in a real-world intrusion against a high-value target. The implications for the ML supply chain are significant: Hugging Face hosts models used in production systems globally, and any tampering with model weights or datasets at scale would be extraordinarily difficult to detect. Security teams building on or consuming resources from model-hosting platforms should review their integrity verification and provenance controls now.

Schrödinger's Feed

ETSI has published TR 104 171, a technical report cataloguing implementation vulnerabilities in Quantum Random Number Generators (QRNGs) — the hardware components increasingly proposed as the entropy foundation for post-quantum cryptographic systems. The report introduces an "Entropy Zero Trust" framework specifically to address side-channel attacks, recognising that even quantum-sourced randomness can leak information through physical implementation flaws. This matters because the security guarantees of post-quantum cryptography are only as strong as the randomness underpinning key generation — a subtly biased or side-channel-leaking QRNG quietly undermines the entire PQC stack. Practitioners evaluating QRNG-backed HSMs or entropy sources for PQC deployments should treat this report as required reading before those procurement decisions are finalised.

/dev/random

This week's finest contribution to the "it's not a bug, it's a philosophical crisis" genre: a 2017 analysis resurfaced this week arguing that CAPTCHAs don't actually verify humanity — they verify American cultural familiarity. The core observation is that image-based CAPTCHAs (identify the fire hydrant, the school bus, the traffic light) disproportionately feature North American street furniture that users from other countries may simply not recognise on sight. A rural user in Southeast Asia failing to identify a US-style fire hydrant isn't failing a humanity test — they're failing a geography test. In 2026, with CAPTCHAs still used as a security primitive for bot detection on everything from login pages to API rate limiting, the underlying assumption that shared visual culture equals humanness deserves a harder look than most threat models give it.