cybr.cx — Daily Intelligence Digest

September 13, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-84869 | ConnectWise ScreenConnect
CISA KEV listed with a remediation deadline of tomorrow, September 14. This flaw allows attackers to transfer files and execute them through active remote sessions without any host confirmation or authorisation — a nightmare scenario for MSPs and any organisation relying on ScreenConnect for remote access. Patch or isolate immediately; this one is being weaponised right now.

⚠️ Actively exploited — CVE-2026-85706 | GitLab CE/EE
Also due September 14. An unauthenticated path traversal in the repository commits API lets attackers read arbitrary files off your GitLab instance — source code, CI secrets, private keys, everything. No credentials required. If you're running a self-managed GitLab instance and haven't patched, assume exposure.

⚠️ Actively exploited — CVE-2026-42016 & CVE-2026-42018 | JFrog Artifactory
Two KEV-listed flaws in Artifactory being exploited in tandem. CVE-2026-42016 is an incorrect authorisation bug that enables privilege escalation by validating token signatures without checking scope. CVE-2026-42018 can leak an internal anonymous-user token to unauthenticated callers even when anonymous access is supposedly disabled. Together, these create a credible unauthenticated-to-admin attack path against one of the most sensitive systems in a software supply chain.

⚠️ Actively exploited — CVE-2026-86060 & CVE-2026-67277 | MikroTik RouterOS
Both due today. CVE-2026-86060 abuses argument delimiter injection to manipulate RouterOS policy masks and escalate privileges. CVE-2026-67277 is a missing-authentication flaw in the btest service enabling kernel memory disclosure and DoS. MikroTik gear is everywhere in ISP infrastructure and enterprise networks — unpatched devices are being actively targeted.

⚠️ Actively exploited — CVE-2026-87491 | Google Chromium V8
An out-of-bounds write in V8 enabling remote code execution via a crafted HTML page — inside the sandbox. Affects Chrome, Edge, and any Chromium-based browser. Patch browsers fleet-wide now; drive-by exploitation is trivially deliverable.

⚠️ Actively exploited — CVE-2026-19490 | Citrix NetScaler ADC/Gateway
Authentication bypass via an alternate path — overdue as of yesterday. When configured as an AAA virtual server or SSL VPN gateway, unauthenticated remote attackers can bypass authentication entirely. Citrix NetScaler is a perennial target for initial access brokers; treat this as critical regardless of your internal risk scoring.

⚠️ Actively exploited — CVE-2026-20079 | Cisco Secure Firewall Management Center / SCC
Authentication bypass on Cisco FMC and Security Cloud Control allowing unauthenticated remote script execution — also overdue. A compromised firewall management plane is about as bad as it gets for network security posture.

⚠️ Actively exploited — CVE-2025-25249 | Fortinet FortiOS, FortiSwitchManager, FortiSASE
Heap-based buffer overflow in multiple Fortinet products exploitable via specially crafted packets to execute arbitrary code. Fortinet edge devices remain a top-tier target for state-sponsored actors; this has been in active exploitation since at least September 9.

⚠️ Actively exploited — CVE-2026-86218 | N-able N-central
Pre-authentication RCE via static code injection in N-able's RMM platform. Like ScreenConnect, N-central sits at the centre of MSP operations and provides an attacker with lateral movement paths to every managed endpoint downstream. High-value target, high urgency.

⚠️ Actively exploited — CVE-2026-81963 & CVE-2026-85880 | Microsoft Windows
Two local privilege escalation flaws — a link-following vulnerability in the Windows Update Stack (CVE-2026-81963) and a heap buffer overflow in ALPC (CVE-2026-85880), both escalating to SYSTEM. Expect these to appear as second-stage payloads paired with initial access techniques.

⚠️ Actively exploited — CVE-2026-75650 | Adobe Commerce / Magento
Template injection in Adobe Commerce and Magento Open Source enabling arbitrary code execution. E-commerce infrastructure running unpatched Magento is a classic skimmer and backdoor target — patch cycles here tend to lag badly.

CVE-2026-78175 | Tutor LMS WordPress Plugin | CVSS 8.8
PHP Object Injection via the withdraw_method_field parameter in an AJAX handler that performs zero capability or role checks — only a nonce stands between an attacker and code execution. Affects all versions up to 4.0.7. If you run any WordPress e-learning infrastructure, update immediately.

CVE-2026-15451 | MemberPress Corporate Accounts (WordPress) | CVSS 8.8
Mass assignment vulnerability in add_sub_account_user passes raw user data including the role key directly to wp_insert_user. Any subscriber-level authenticated user can self-promote to administrator. Affects versions up to 1.5.39.

CVE-2026-90553 | vLLM < 0.28.0 | CVSS 7.8
The LlavaOnevision2 processor loader ignores trust_remote_code=False and will execute arbitrary Python in processing_llava_onevision2.py from a malicious model repository. If your organisation is serving or experimenting with remote models via vLLM, this is a supply chain RCE waiting to happen.

Headline News

AI Agents Weaponised Against Open Source Package Infrastructure

Threat actors have used AI agent pipelines — specifically instances of Claude — to autonomously analyse and extract secrets from approximately 1.8 million Android applications at scale. According to Anthropic's own disclosure, multiple threat groups were involved, including financially motivated actors and state-sponsored clusters attributed to Russia and China. The technique involved automating what would typically be manual reverse engineering: decompiling APKs, identifying hardcoded API keys, OAuth tokens, cloud credentials, and private endpoints, then exfiltrating findings programmatically. The scale is what makes this significant — operations that previously required large teams of analysts can now be parallelised through AI agents at a fraction of the cost. Separately, OpenAI agents were reportedly leveraged in a targeted attack against RubyGems, demonstrating that the same capability is being applied against package registry infrastructure, a critical chokepoint in software supply chains. Security teams need to treat AI-assisted reconnaissance as a first-class threat model, not a future concern.

Florida DMV Breach Compounds Driver's Licence Data Crisis

Florida's Department of Highway Safety and Motor Vehicles has confirmed it is investigating a data breach — the second major incident involving Americans' driver's licence records this month. The timing is notable: the breach occurred shortly after a separate large-scale driver's licence compromise, raising the question of whether attackers are systematically targeting motor vehicle administration infrastructure as a high-value source of government-issued identity data. Driver's licence records are particularly damaging because they combine full legal name, address, date of birth, and a document number that unlocks a range of downstream fraud and identity theft scenarios. For practitioners, this is a reminder that state government agencies often carry large, sensitive datasets with security postures that lag behind their data sensitivity. Organisations that rely on driver's licence data for identity verification should monitor for signs of synthetic identity fraud in the weeks ahead.

Liquid Network Exploited for Largest Bitcoin Sidechain Hack of 2026

An attacker minted 3,998 L-BTC on the Liquid Network — Blockstream's Bitcoin sidechain — by exploiting a cryptographic verification flaw in the network's federation mechanism. Analysis from SlowMist indicates the exploit targeted weaknesses in how the sidechain validates cryptographic proofs, allowing the attacker to mint native sidechain assets without the corresponding Bitcoin collateral. At current valuations this represents a significant theft, and it exposes a systemic issue: many sidechain and layer-2 designs inherit assumptions about cryptographic verification that have not been rigorously audited at the implementation level. For security practitioners working in blockchain or fintech environments, the incident underscores that peg-in/peg-out mechanisms and cross-chain bridges remain among the highest-risk attack surfaces in the ecosystem. Immediate audits of cryptographic verification pathways in any sidechain deployment are warranted.

Schrödinger's Feed

India's National Quantum Mission has formalised a collaboration between QClairvoyance Quantum Labs and the IITM–C-DOT Samgnya Foundation, with post-quantum cryptography listed as an explicit workstream alongside quantum computing and AI. India joining the growing roster of nations with state-backed PQC development programs matters geopolitically — it broadens the field of entities building independent cryptographic infrastructure ahead of the quantum threat horizon. With NIST's PQC standards now finalised and migration timelines hardening, state-level investment signals that the window for organisations to begin crypto-agility planning is closing faster than many procurement cycles allow. Practitioners should be asking their vendors today what their PQC migration roadmap looks like — because their adversaries' governments already are.

/dev/random

This week's oddity: Nvidia has apparently become so central to the AI economy that serious financial analysis now compares the company to a central bank — the entity that controls the reserve asset everyone else needs to function. The analogy holds up uncomfortably well: GPU allocation decisions shape which AI projects live or die, much like interest rate decisions shape which businesses can borrow. From a security angle, this level of concentration in a single vendor's hardware introduces a systemic risk that no threat model adequately accounts for — a supply chain disruption, sanctions event, or critical firmware vulnerability in Nvidia's ecosystem would propagate across essentially every major AI deployment simultaneously. The "central bank of AI" framing is catchy, but "single point of failure for global AI infrastructure" is the version that should be in your risk register.