Skip to content
RIZON
Back to Threat Briefing
Threat Briefing — Edition #4

Threat Briefing Q4 2026 (November) — SharePoint Under Siege, Edge Zero-Day Chains, and AI Infrastructure as Initial Access

Orizon TeamNovember 17, 202613 min readThreat Intelligence

Four patterns defined the threat landscape between June and mid-November 2026. On-premises Microsoft SharePoint went through a sustained run of exploitation. Edge appliances kept failing, in one case to a chain of zero-days that left few forensic traces. The tools that administer everything else (remote monitoring and management, CI/CD and virtualisation consoles) were turned into entry points. And self-hosted AI infrastructure moved from an emerging risk to a routine initial-access target.

The regulatory backdrop shifted too. Since September 11, 2026, manufacturers of products with digital elements sold in the EU must report actively exploited vulnerabilities under the Cyber Resilience Act. In the United States, CISA replaced its long-standing KEV directive with one that ties many remediation deadlines to three days and a forensic check. For European organisations under NIS2 and DORA, the practical message is the same: an exploited vulnerability on an internet-facing system is now an incident-response question, not only a patching one.

All CVSS scores below are the scores published in the NIST National Vulnerability Database (from the vendor or other CNA, from NIST, or both where they differ). KEV dates and ransomware flags come from the CISA Known Exploited Vulnerabilities catalog as of its November 13, 2026 release.


At a Glance

MetricDetail
174Vulnerabilities added to CISA's KEV catalog between June 1 and November 13, 2026
9Of those, flagged by CISA as known to be used in ransomware campaigns
6On-premises SharePoint Server flaws added to KEV between July 1 and September 25
3 flaws, 0 web shellsIvanti Connect Secure zero-day chain: root on a fully patched appliance, commands passed through the exploit itself
128 of 164KEV additions since CISA's BOD 26-04 (June 10) carrying a three-day remediation deadline
9AI and ML platform flaws (LiteLLM, Langflow, Ray, MLflow, Kestra, GPUStack) added to KEV in the period
September 11, 2026CRA Article 14 reporting obligations apply: 24-hour early warning to the coordinating CSIRT and ENISA

1. On-Premises SharePoint: A Six-CVE Exploitation Series

Six SharePoint Server vulnerabilities entered the KEV catalog in under three months. One is linked to ransomware.

Between July 1 and September 25, 2026, CISA added six on-premises SharePoint Server vulnerabilities to KEV. CERT-EU advisory 2026-009, issued in July and covering the first four, describes CVE-2026-50522 as part of an ongoing series of actively exploited flaws in on-premises SharePoint, and its recommendations go beyond patching.

CVETypeCVSS (NVD)KEV addedRansomware use (KEV)
CVE-2026-45659Deserialisation, authenticated RCE8.8July 1Known
CVE-2026-56164Missing authentication, privilege escalation9.8 (NIST) / 5.3 (Microsoft)July 14Unknown
CVE-2026-58644Deserialisation, unauthenticated RCE9.8July 16Unknown
CVE-2026-50522Deserialisation, RCE9.8July 22Unknown
CVE-2026-55040Weak authentication, security feature bypass9.1August 18Unknown
CVE-2026-65660Code injection, authenticated RCE8.8September 25Unknown

Two details matter for prioritisation:

  • Vendor preconditions were challenged in practice. For CVE-2026-50522, CERT-EU notes that Microsoft lists some authentication as a precondition, but that recent findings suggest exploitation may not require it. CERT-EU also records that watchTowr identified proof-of-concept exploit code on July 20 and then observed active exploitation.
  • Severity ratings disagree. Microsoft scores CVE-2026-56164 at 5.3. NIST scores the same flaw 9.8. A triage process keyed to the vendor score alone would have queued an exploited privilege-escalation flaw as medium.

Why This Matters

On-premises SharePoint often holds the documents that NIS2 and DORA audits rely on: policies, risk registers, contracts and board papers. It also sits close to Active Directory. CERT-EU's guidance is to update, rotate credentials for any vulnerable asset that was exposed to the internet, run a compromise assessment, and reconsider exposing SharePoint Server directly to the internet at all. Patching alone does not evict an attacker who arrived before the patch.

What to Do

ActionFramework Reference
Inventory every on-premises SharePoint farm and confirm all servers in each farm carry the latest cumulative updateNIS2 Art. 21(2)(e), ISO 27001 (8.8), NIST 800-53 SI-2
Remove direct internet exposure; publish only through an authenticating reverse proxy or move workloads to managed servicesNIS2 Art. 21(2)(a), ISO 27001 (8.20, 8.22), NIST 800-53 SC-7
After patching, rotate credentials and keys on previously exposed servers and run a compromise assessmentDORA Art. 11, ISO 27001 (5.26), NIST CSF RS.MA
Triage on exploitation status (KEV, CERT-EU) as well as CVSS; treat a vendor/NIST score disagreement as a reason to escalateDORA Art. 9, ISO 27001 (5.7, 8.8), NIST CSF ID.RA-01

2. The Edge Keeps Breaking — Now in Chains

On a fully patched Ivanti appliance, an attacker chained three zero-days to root, deployed no web shell, and erased its own log entries.

Volexity reported on October 13 that a threat actor it tracks as UTA0576 had repeatedly exploited Ivanti Connect Secure (ICS) from October 3, 2026, compromising an appliance running 22.8R2.4, then fully patched. The chain combined an authentication bypass, a file-write and race-condition RCE, and a privilege escalation in how the dsinvoked broker trusts cached configuration data. Ivanti released fixes on October 13 under CVE-2026-104935, which Volexity said tracks both the authentication bypass and the RCE. Volexity reports that the privilege escalation (CVE-2026-111357) was patched on October 29. On October 30, CISA added that flaw and a predictable-filename upload flaw (CVE-2026-118527) to KEV, noting that each can be chained with the other two Ivanti CVEs.

The tradecraft is the concern:

  • No persistence on the device. Commands travelled inside the exploit requests themselves. Volexity found "no backdoor or web shell deployed".
  • Anti-forensics. The payload selectively deleted recent entries from the ICS debug log and the Linux audit log.
  • Integrity checks can miss it. Volexity reports that Ivanti's Integrity Checker Tool "does not detect any modifications" from this chain. CISA's alert therefore recommends network and memory forensics, a factory reset or rebuild, and a reset of every account password, certificate and API key held on the appliance.
  • Motive. Volexity assesses with low confidence that UTA0576 is criminally rather than espionage-motivated. CISA reports credential theft and data exfiltration.

Ivanti was not alone. The period's exploited edge and security-management flaws included:

CVEProductCVSS (NVD)KEV addedRansomware use (KEV)
CVE-2026-50751Check Point Remote Access VPN (IKEv1) — authentication bypass9.3June 8Known
CVE-2026-15409SonicWall SMA1000 — unauthenticated SSRF10.0July 14Known
CVE-2026-15410SonicWall SMA1000 — post-auth code injection7.2July 14Known
CVE-2026-20316Cisco Secure Firewall Management Center — static credentials5.3July 29Known
CVE-2026-19490Citrix NetScaler ADC/Gateway — authentication bypass9.8 (NIST) / 9.3 (v4.0, Citrix)September 9Unknown
CVE-2026-88771Citrix NetScaler ADC/Gateway — unauthenticated RCE9.8 (NIST) / 9.5 (v4.0, Citrix)September 27Unknown
CVE-2026-88772Citrix NetScaler ADC/Gateway — memory overflow (DTLS)8.1 (NIST) / 9.5 (v4.0, Citrix)September 27Unknown
CVE-2026-104935Ivanti Connect Secure — auth bypass + RCE8.6October 13Unknown
CVE-2026-111357Ivanti Connect Secure — privilege escalation to root7.0October 30Unknown
CVE-2026-118527Ivanti Connect Secure — predictable file names4.3October 30Unknown

Three observations:

  • Check Point disclosed on June 8 that exploitation of CVE-2026-50751 dated back to May 7. It affected a few dozen organisations, and in one case the post-compromise activity was linked to a Qilin ransomware affiliate. The flaw sits in the deprecated IKEv1 key exchange, and several affected versions are already past end of support.
  • Citrix NetScaler produced three CERT-EU critical advisories in the period: 2026-010 (August), 2026-014 (September, two flaws confirmed exploited) and 2026-020 (November 12). The November bulletin covers 28 flaws. Citrix reports no exploitation of them so far, but given the appliance's history it should be patched on the same urgent schedule.
  • Low scores, real ransomware. Cisco FMC CVE-2026-20316 scores 5.3 and Ivanti CVE-2026-118527 scores 4.3, yet the first is flagged for ransomware use and KEV notes that the second can be chained with the two other Ivanti flaws exploited in October. CVSS measures a flaw on its own. It does not measure what the flaw enables when chained with others.

Why This Matters

The ENISA Threat Landscape 2026 (published September 22, covering 2025) found that in the unauthorised-access incidents where an intrusion vector could be identified (only 5% of them), 60% involved exploiting a vulnerability. It also found that 73% of targeted organisations were NIS2 essential or important entities. VPN gateways and firewall managers are where that exploitation starts. When the attacker leaves no file on the device and edits its logs, "patched and integrity-checked" is not evidence of "not compromised".

What to Do

ActionFramework Reference
For any edge device exploited before you patched it, treat the event as a suspected incident: memory/network forensics, rebuild, and rotate every credential, certificate and key it heldNIS2 Art. 21(2)(b), Art. 23; DORA Art. 17, 19; ISO 27001 (5.24–5.26)
Send edge-appliance logs off-box in real time so on-device log tampering cannot erase the recordISO 27001 (8.15, 8.16), NIST 800-53 AU-9, NIST CSF DE.CM-01
Retire deprecated protocols (IKEv1) and end-of-support appliance versions; track them as risk-register items until removedNIS2 Art. 21(2)(e), ISO 27001 (8.9), CIS Controls 4
Build exploit chains into vulnerability triage: a low-CVSS flaw on an exposed edge device is high priority if it appears in KEVDORA Art. 9, ISO 27001 (8.8), NIST 800-53 RA-5

3. The Admin Plane Is the Target

The software that manages, builds and deploys everything else was a steady source of exploited, critical flaws.

Remote monitoring and management (RMM), remote support, CI/CD and virtualisation consoles hold privileged, network-wide reach by design. One exploited flaw in any of them gives an attacker that reach. In this period, CISA added at least ten such flaws to KEV, eight of them rated critical by NVD or the vendor:

CVEProductCVSS (NVD)KEV addedRansomware use (KEV)
CVE-2026-48558SimpleHelp — OIDC token signature not verified10.0June 29Unknown
CVE-2026-18577N-able N-central — auth bypass (incomplete fix of CVE-2026-18556)8.1 (NIST) / 8.2 (v4.0, N-able)August 3Unknown
CVE-2026-18556N-able N-central — authentication bypass7.4 (NIST) / 8.2 (v4.0, N-able)August 4Unknown
CVE-2026-63077JetBrains TeamCity — unauthenticated RCE via agent polling9.8August 5Known
CVE-2026-59310VMware vCenter — syslog path traversal to code execution9.8August 18Known
CVE-2026-82329JFrog Artifactory — unauthenticated admin under default configuration9.8September 2Unknown
CVE-2026-86218N-able N-central — pre-authentication RCE9.8 (NIST) / 10.0 (v4.0, N-able)September 8Unknown
CVE-2026-84869ConnectWise ScreenConnect client — unauthorised file transfer and execution in a session9.9September 11Unknown
CVE-2026-85706GitLab CE/EE — unauthenticated arbitrary file read10.0September 11Unknown
CVE-2026-102764BeyondTrust Remote Support — authentication bypass by spoofing9.8 (NIST) / 9.3 (v4.0, BeyondTrust)October 16Unknown

The N-able N-central sequence shows how such flaws play out. An authentication bypass was followed by an incomplete fix, then a separate pre-authentication RCE. All three were added to KEV between August 3 and September 8. N-able patched its hosted instances itself; self-hosted customers had to upgrade. When it posted the hotfix notice on September 6, N-able said it had no confirmation of exploitation. CISA added the flaw to KEV on September 8.

JFrog Artifactory added three further KEV entries in the period (CVE-2026-66384, CVE-2026-42016, CVE-2026-42018). An artifact repository serves the packages your builds trust, so it is a supply-chain control in its own right.

Why This Matters

For an MSP, an RMM compromise reaches every customer. For a financial entity under DORA, a compromised RMM or CI/CD provider is an ICT third-party risk under Article 28, not only an internal IT issue. NIS2 Article 21(2)(d) requires supply chain security that covers "relationships between each entity and its direct suppliers or service providers". Article 21(3) adds that entities must consider each supplier's vulnerabilities and the quality of its secure development procedures.

What to Do

ActionFramework Reference
Keep RMM, remote-support, CI/CD and virtualisation management interfaces off the public internet; require VPN/ZTNA with phishing-resistant MFANIS2 Art. 21(2)(j), ISO 27001 (8.5, 8.20), NIST CSF PR.AA-03
Maintain an inventory of admin-plane tools, including those operated by your MSPs, with a named patch owner and emergency SLA for eachDORA Art. 28, NIS2 Art. 21(2)(d), ISO 27001 (5.9, 5.19–5.22)
Contractually require suppliers to notify you of exploited vulnerabilities in tools that touch your environmentDORA Art. 28–30, ISO 27001 (5.20), NIST CSF GV.SC
Treat build servers and artifact repositories as tier-0 assets: least privilege, isolated networks, integrity checks on artifactsISO 27001 (8.2, 8.25), NIST 800-53 AC-6, SR-3

4. AI Infrastructure Has Become an Initial-Access Target

Edition 003 covered a single Langflow flaw exploited within 20 hours. Since June, nine AI-platform flaws have entered KEV.

Self-hosted AI tooling (LLM gateways, agent builders, workflow orchestrators and GPU cluster managers) now appears regularly in CISA's exploited catalogue:

CVEProductCVSS (NVD)KEV added
CVE-2026-42271LiteLLM — command injection via MCP test endpoints (low-privilege keys suffice)8.8June 8
CVE-2026-55255Langflow — IDOR lets a user run other users' flows8.4July 7
CVE-2026-0770Langflow — unauthenticated RCE via exec_globals9.8 (CVSS 3.0, ZDI)July 21
CVE-2026-9198Langflow — auto_login superuser token + code validation endpoint = unauthenticated RCE9.8August 4
CVE-2025-62593Ray — code injection reachable via Firefox and Safari for developers running Ray locally8.8 (NIST) / 9.4 (v4.0, GitHub)August 17
CVE-2026-64849MLflow — unauthenticated SSRF to internal and cloud metadata services9.3August 19
CVE-2026-49869Kestra OSS — auth bypass leading to unauthenticated workflow execution10.0September 2
CVE-2026-59822LiteLLM — MCP endpoint accepts an arbitrary Bearer token8.2 (NIST) / 8.8 (v4.0, GitHub)September 2
CVE-2026-102895GPUStack — default loopback trust gives unauthenticated admin9.8November 9

None of these flags is set to "Known" ransomware use in KEV, but the access they give is high value. An LLM gateway holds provider API keys and sees every prompt and response. MLflow's SSRF reaches cloud metadata services. GPUStack's flaw exposes worker and cloud-provider secrets.

Wiz published honeypot results on August 27 from 90 days of telemetry across AI services. Attackers exploited the LiteLLM MCP flaw with Bearer tokens as trivial as a single character, tried LiteLLM's default master key sk-1234, queried the running process's state to extract the master key, and deployed cryptominers on Langflow and Node-RED hosts.

A recurring root cause is insecure defaults. Langflow's auto_login handed superuser tokens to any network caller on default deployments. GPUStack trusted loopback-looking requests as the administrator by default. Kestra exempted any path ending in /configs from authentication.

Why This Matters

AI platforms are often deployed by data or product teams outside the normal change, exposure-review and patch processes. That puts them outside the asset inventory that NIS2 Article 21(2)(i) and ISO 27001 control 5.9 expect. Under the EU AI Act, high-risk AI systems must meet accuracy, robustness and cybersecurity requirements (Article 15). The Annex III high-risk deadlines were postponed to December 2, 2027 by Regulation (EU) 2026/1744. The infrastructure exposure, though, exists today, whatever the AI Act timetable.

What to Do

ActionFramework Reference
Discover and inventory self-hosted AI services (gateways, agent builders, notebooks, model servers) across cloud accountsNIS2 Art. 21(2)(i), ISO 27001 (5.9), NIST CSF ID.AM-02
Never expose AI admin or workflow UIs to the internet; disable auto-login and default keys; enforce authentication on every APIISO 27001 (8.5, 8.9), CIS Controls 4, NIST 800-53 CM-7
Store LLM provider keys in a secrets manager with scoped, rotatable credentials, and alert on unusual spendISO 27001 (5.17, 8.24), NIST 800-53 IA-5
Block AI hosts from reaching cloud metadata endpoints; enforce IMDSv2 or equivalentISO 27001 (8.20, 8.22), NIST 800-53 SC-7

Cross-Cutting Trend: "Patched" Is No Longer the End State

Every theme in this edition points to the same operational shift: remediation now includes a compromise check.

  • CISA BOD 26-04 (issued June 10, 2026) supersedes and revokes BOD 22-01, the directive that created the KEV catalog. For entries it marks "& forensic triage", it sets a three-day window covering both remediation and forensic triage: agencies must check whether the asset was already compromised. Of the 164 KEV entries added from June 10 to November 13, 128 carried a three-day due date. The directive binds US federal civilian agencies only, but KEV deadlines are widely used as a prioritisation signal elsewhere.
  • CERT-EU's SharePoint advisory pairs patching with credential rotation and a compromise assessment; its September NetScaler advisory also calls for a compromise assessment of internet-facing appliances.
  • Volexity and CISA advise that integrity tooling may not detect the Ivanti chain, so forensics and a rebuild are required.

For NIS2 entities this matters for Article 23. The 24-hour early warning and 72-hour incident notification are triggered by a significant incident, not by a vulnerability. An entity can only meet those deadlines if it finds out quickly whether a flaw on its systems was exploited before the patch landed.

Law-enforcement pressure on the ecosystem

Operation Endgame continued with an action week from June 15 to 19, 2026. Authorities disrupted the SocGholish, StealC and Amadey malware services, took down 326 servers and 142 domains, and recovered 27 million compromised data sets. Eurojust describes all three as initial-access malware and cites ransomware installation as one use of SocGholish access, so the disruption bears directly on ransomware risk.


Regulatory Follow-Up Since Edition 003

Item flagged in AprilStatus, November 2026
CRA vulnerability reportingLive since September 11, 2026 (Art. 71(2)). Under Art. 14, manufacturers notify actively exploited vulnerabilities to the coordinating CSIRT and ENISA via the single reporting platform: early warning within 24 hours, notification within 72 hours, final report within 14 days of a fix being available. See our Compliance Q&A on CRA reporting.
EU AI Act high-risk obligations (August 2, 2026)Annex III high-risk obligations postponed to December 2, 2027 by Regulation (EU) 2026/1744.
NIS2 in the NetherlandsThe Cyberbeveiligingswet (Cbw) has been in force since August 15, 2026.

The CRA point is directly relevant to this edition. Many of the vendors whose products appear above sell into the EU. From September 11, an actively exploited vulnerability in such a product triggers the manufacturer's 24-hour duty to notify. If you build products with digital elements, your own exposure to the edge, admin-plane and AI flaws above may now trigger a CRA report as well as a NIS2 one.


What to Watch Next Quarter

  • Citrix NetScaler (CERT-EU 2026-020): 28 flaws disclosed November 12 with no exploitation reported yet. Given the appliance's record this year, close the window before exploitation is reported.
  • Ivanti Connect Secure 22.8 end of life: January 31, 2027 per CISA. Plan migrations now rather than under incident pressure.
  • On-premises SharePoint: six exploited flaws in three months suggest more to come. Consider whether exposure to the internet is still justified.
  • AI infrastructure: expect further KEV entries for LLM gateways and agent builders. Put AI tooling under the same exposure management as any other internet-facing service.
  • CRA reporting in practice: the first months of notifications through ENISA's single reporting platform will show how manufacturers interpret "actively exploited".

Threat Briefing is a quarterly series providing strategic threat analysis for European mid-market organisations. Each edition connects current threats to regulatory obligations and provides actionable recommendations mapped to recognised frameworks. Subscribe to our RSS feed to receive each edition as it publishes.