This Week in Exposure: 2,906 New CVEs, an AI Agent Breach, and the Five Fixes That Matter

This Week in Exposure: 2,906 New CVEs, an AI Agent Breach, and the Five Fixes That Matter

2,906 new CVEs, 10 exploited and an AI agent breach at DIVD. The fixes financial firms and SaaS teams need now, with versions and steps.

An automated AI agent breached the Dutch institute that warns the world about vulnerabilities. In the same seven days, 2,906 new CVEs were published and 10 flaws were confirmed as exploited in the wild. If you run a financial services firm or a software platform, five of them need action this week. Here is which ones, who is exposed, and what to do.

The hackers got hacked, by an agent

On 24 September, DIVD, the Dutch Institute for Vulnerability Disclosure, announced it had been breached. DIVD is the volunteer organisation that scans the internet for exposed systems and warns their owners. It has been a registered CVE Numbering Authority since 2022.

DIVD has since published how it got in (DIVD-2026-00014, DIVD-2026-00015). On 21 September the attacker exploited two previously unknown flaws in Zammad, the open source helpdesk DIVD ran: CVE-2026-102489, remote code execution, and CVE-2026-102490, escalation to root. An autonomous agent chained them. In DIVD's words, the pair let the attackers "hijack sessions, run code remotely, and escalate privileges from the Zammad user to root, in seconds, due to the agentic part of this hack" (BleepingComputer). DIVD blocked access on 22 September, reported the flaws to Zammad on 24 September, and from 26 September scanned the internet for other exposed Zammad servers and warned their owners. Network segmentation kept the attacker out of deeper parts of the network.

If you run Zammad, or your IT or support provider runs it for you: upgrade to Zammad 7 or take it offline, as DIVD advises, and secure your application and network logs before you update, as the Dutch NCSC advises (Security.NL). A helpdesk faces the internet by design and holds customer data by design: tickets, attachments, and the passwords people should never have pasted into them.

That is the lesson for everyone else. An attacker no longer needs a skilled team working your network by hand. A cheap, badly configured agent that never sleeps is enough to find the one system you did not patch. Which makes the question for this week simple: which of your systems are on the list below?

The week in numbers

Nobody patches 2,906 things in a week, and nobody needs to. The job is finding the few that are being exploited, are reachable from the internet, and sit in your estate or your provider's. Four of this week's exploited flaws were in remote access gateways and network controllers, the systems that face the internet by design. All four were attacked before their vendors had a fix.

Your 60 second triage

Do not run any of these yourself? Your IT provider, hosting partner or platform vendor may. Section 4 has the three questions to send them.

1. Citrix NetScaler: attacked before the patch existed

What happened. On 27 September Citrix disclosed eight flaws in NetScaler ADC and NetScaler Gateway (CTX697096). Two of them, CVE-2026-88771 and CVE-2026-88772 (CVSS 9.5), let an attacker run code without logging in, and both were exploited before Citrix published a fix. GreyNoise logged its first exploitation attempt on 24 September, three days before the bulletin (The Hacker News). Palo Alto Networks Unit 42 counted 50,277 exposed instances. CVE-2026-88771 works against default configurations. CVE-2026-88772 needs DTLS, which is on by default for VPN virtual servers.

Who should act now.

• Wealth managers, family offices, trust and corporate service providers and fund administrators whose staff reach client systems through a Citrix Gateway login page.
• Any firm whose IT provider hosts its desktops or applications on Citrix. You may never see the appliance, but your client data sits behind it.
• SaaS and platform companies that use NetScaler ADC as the load balancer in front of production.

Why it will not wait. A remote access gateway sits on the internet by design, and one crafted request can be enough. And if you patched NetScaler in August for the earlier flaw CVE-2026-19490, you are not covered: the August builds are exposed to these two, and only the builds released on 27 September fix them (watchTowr). This is exactly the kind of target automated scanning finds first.

Do this now.

1. Upgrade to NetScaler 14.1-73.37 or later, 13.1-64.23 or later, or 14.1-73.37 FIPS or later.
2. Treat any appliance that was reachable since 24 September as possibly compromised, and follow Citrix's steps for a suspected compromise.
3. If your IT provider runs Citrix for you, ask them in writing which build you are on and when it was installed.

Same week, same door. Check Point VPN gateways joined CISA's exploited list on 22 September (CVE-2026-85102), after the Dutch NCSC warned in mid September that exploitation was likely (Security Affairs). If your firewall is Check Point, confirm you are on R82.20 or have the LivePatch or Jumbo Hotfix Check Point released on 9 September.

2. SharePoint on your own servers: six weeks was too slow

What happened. Microsoft fixed CVE-2026-65660 (CVSS 8.8) in its August 2026 security updates. On 25 September Microsoft confirmed attacks, and CISA added it to the exploited list the same day (SecurityWeek). Early warning firm Previdian saw exploitation attempts on 24 September and attempts to plant a webshell, a hidden backdoor page, on 25 September.

Who should act now.

• Financial firms that keep client files, KYC records, board packs or compliance evidence on SharePoint Server hosted in their own server room or by their IT provider. Microsoft's fixes cover SharePoint Server Subscription Edition, 2019 and 2016 (fixed builds).
• Software companies still running SharePoint Server as an internal intranet or document store.

Why it will not wait. The flaw needs a login, but only an ordinary low privileged one. Infostealer malware and password spraying supply those at scale, and password spraying is exactly what the agent in the DIVD breach was doing. From one stolen account to code running on the server that holds your client files is a single step. The patch was available for six weeks. That was not fast enough.

Do this now.

1. Install the August 2026 SharePoint security update for your version, and confirm the build number afterwards.
2. Disable dormant and shared accounts, and require multi factor authentication on every SharePoint login.
3. Check your SharePoint servers for unexpected new files or web pages created since 24 September.

3. WSO2: forged tokens against the API layer

What happened. CVE-2026-5430 (CVSS 9.8) lets attackers bypass authentication on WSO2 API Manager and its gateway components with forged login tokens (JWTs), which can hand them administrator accounts. watchTowr captured forged tokens aimed at the flaw from 13 September, before technical details were public. CISA added it to the exploited list on 24 September, alongside an Adobe Commerce flaw that lets attackers switch into another customer's account (The Hacker News).

Who should act now.

• SaaS and fintech companies that publish or protect their APIs with WSO2 API Manager 4.1.0 to 4.6.0, or WSO2 API Control Plane, Traffic Manager or Universal Gateway 4.5.0 and 4.6.0 (affected versions).
• Financial firms whose platform provider, payment partner or integration vendor exposes APIs through WSO2. You will not see it in your own estate. Ask.
• Online stores on Adobe Commerce or Magento 2.4.4 to 2.4.9 without the August 2026 update APSB26-92.

Why it will not wait. Admin on an API gateway means control over every API behind it: who can call them, what they return, which keys are valid. Attackers were working this flaw for 11 days before it reached CISA's list. If you waited for the list, you started late.

Do this now.

1. Apply the WSO2 update level that fixes CVE-2026-5430 for your product and version.
2. Rotate token signing keys and administrator credentials after patching.
3. Review administrator accounts and API keys created or changed since 13 September.

4 and 5. F5 and VeloCloud: the infrastructure your provider probably runs

Most small and midsize financial firms, and plenty of scaling software companies, do not run F5 or VeloCloud themselves. Their managed service provider, hosting partner or network provider does. Both flaws below were exploited before a patch existed, and both sit on systems that control access for everyone behind them.

F5 BIG-IP APM, CVE-2026-94127 (CVSS 9.8). An attacker can run code on the device without logging in, but only when BIG-IP APM acts as an OAuth Authorization Server, meaning an access policy and an OAuth profile on the same virtual server. Affected versions are 17.1.0 to 17.1.3, 17.5.0 to 17.5.1 and 21.1.0 (CERT-EU). Shadowserver sees more than 14,700 internet facing BIG-IP APM fingerprints (Security Affairs). F5 has hotfixes and an iRule mitigation for systems that cannot be patched straight away (Rapid7).

VeloCloud Orchestrator, CVE-2026-93952 (CVSS 10.0). The orchestrator is the control centre of a VeloCloud SD-WAN and holds the configuration of every site it manages. On premises orchestrators that authenticate their edge devices with certificates are exposed. Fixed builds are 5.2.3.16 and 6.4.2.8; some other release trains were still waiting on fixes at disclosure (SecurityWeek, Arista advisory). Arista advises incident response after upgrading, not just the upgrade.

Who should act now. Any firm that outsources its network, remote access or single sign on. Logistics and retail businesses running many sites over SD-WAN in particular. For financial entities under DORA, managing ICT third party risk is a legal duty, and this is a live test of it.

Three questions to send your provider this week.

1. Do you run F5 BIG-IP APM as an OAuth server, or VeloCloud Orchestrator on premises, anywhere that touches our service?
2. Are those systems on the fixed builds, and since what date?
3. Have you checked them for signs of compromise, as F5 and Arista both advise?

And when a vendor tells you to switch off. On 25 September Kiteworks, a secure file sharing platform used by financial institutions and government bodies, asked customers to shut down their servers early on Saturday 26 September, from 04:00 to 10:00 Central European time. It cited credible threat intelligence from federal authorities, said it had found no evidence of compromise, pointed customers to version 9.5.1, and lifted the recommendation on 27 September (BleepingComputer, The Hacker News). Nothing broke. But if your client file exchange went dark at a few hours' notice, would your team know what to do instead? Under DORA, that is the business continuity question in real life.

What DIVD got right in the first 24 hours

DIVD detected suspicious activity, isolated its infrastructure, brought in a third party incident response team, and reported to the Dutch data protection authority (Autoriteit Persoonsgegevens) and the National Cyber Security Centre on the day it went public. It also told the world when its next update would come, and kept that promise. Few organisations would manage that sequence cleanly on their first real incident.

It is also, almost exactly, the drill regulated firms in the EU are expected to run:

Agents do not keep office hours. If your first 24 hours depend on one person reading an alert on Monday morning, the clock has already run out. Write down who classifies an incident, who calls the regulator, and who calls your IT provider, before you need to.

Five takeaways

1. Volume is noise. 2,906 new CVEs, 10 exploited, 5 that matter to most firms. Prioritise by exploitation and exposure, not by the score alone.
2. The edge gets hit first. Every flaw attacked before a patch existed this week sat on a gateway or controller facing the internet.
3. Patch lag is the real risk. SharePoint was fixed in August and exploited in September. A monthly patch cycle now loses races it used to win.
4. Your exposure includes your providers'. If you outsource IT, their unpatched F5 or VeloCloud is your incident, and under DORA your reporting obligation.
5. Attackers now run agents. Sloppy, cheap and tireless. The organisations that close the obvious doors first are the ones the agents move past.

Where to start

If you want to see what an automated attacker sees when it looks at your organisation, get a free exposure report. We map your internet facing systems, check them against what is being exploited this week, and show you the few paths that actually reach your data.

If you already know something on this list touches your estate, or your provider's, talk to one of our security experts this week rather than next.

Sources

• DIVD CSIRT: It was a matter of when, not if, 24 September 2026
• DIVD case DIVD-2026-00014: the breach and DIVD-2026-00015: the Zammad vulnerabilities, updated 30 September 2026
• BleepingComputer: Automated AI agent used to breach cybersecurity nonprofit DIVD, 29 September 2026
• BleepingComputer: DIVD says Zammad zero days enabled AI driven network breach, 30 September 2026
• Security.NL: Zammad zero days used in DIVD hack, 1 October 2026
• NVD CVE API, CVEs published 22 to 28 September 2026
• CISA Known Exploited Vulnerabilities catalog and alerts of 22 September, 24 September, 25 September and 27 September
• Citrix bulletin CTX697096 and CTX694799
• Unit 42: NetScaler zero days exploited
• watchTowr: NetScaler zero day FAQ
• The Hacker News: CISA says attackers are exploiting two critical NetScaler flaws
• The Cyber Express: NetScaler flaws exploited
• SecurityWeek: SharePoint CVE-2026-65660 now exploited
• HOL: CVE-2026-65660 fixed builds
• The Hacker News: WSO2 and Adobe Commerce flaws exploited
• AiCybr: WSO2 CVE-2026-5430 affected versions
• Rapid7: CVE-2026-94127 in F5 BIG-IP APM
• CERT-EU advisory 2026-013
• Security Affairs: F5 BIG-IP APM zero day
• Security Affairs: CISA adds Check Point, VeloCloud and F5 flaws
• Arista Security Advisory 0183
• SecurityWeek: Arista urges immediate patching of VCO zero day
• BleepingComputer: Kiteworks urges server shutdown
• The Hacker News: Kiteworks urges customers to shut down systems
• AFM: DORA ICT related incidents
• NIS2 Directive (EU) 2022/2555
• GDPR Article 33