Issue #015 · Week of September 27, 2026
This Week’s Verdict
The edge devices are on fire again (Citrix, F5, VeloCloud, MikroTik, Check Point), the extortion crews have started eating each other, and the AI agents everyone rushed into production keep wandering into places they were never authorized to be. If you were hoping the industry had learned something over the summer, I have bad news and a KEV catalog to show you.
The Breaches
The most entertaining story of the week is also the most instructive: ShinyHunters defaced Clop’s data leak site, moving it to a fresh Tor address and claiming to have stolen Clop’s victim data. They pulled it off using an unauthenticated path traversal flaw in Grav CMS, because apparently ransomware operators patch their own infrastructure about as diligently as their victims. If ShinyHunters actually has that data, every organization that quietly paid Clop is now exposed to a second round of extortion. Paying the ransom bought you nothing but a receipt.
ShinyHunters also claims to have breached the FBI, allegedly stealing data on agents and job applicants, which they frame as revenge for a bureau warning about their tactics. Treat the claim with appropriate skepticism, but the theatrics are the point.
Elsewhere, suspected North Korean actors lifted $351.6 million from crypto exchange Bitget via a backend compromise. Lazarus continues to be the most reliably funded threat actor on earth, and it is funded by the rest of us. And CrowdSec, a security firm, lost the contents of 170 private GitHub repositories to the Shai-Hulud worm via an OAuth token stolen from a former employee’s machine. When the people selling defense get caught by npm supply-chain malware, the lesson generalizes.
Finally, a U.S. Army soldier got 70 months for the AT&T and Verizon metadata thefts affecting over 100 million customers, and the Rydox marketplace admin pleaded guilty, facing up to 22 years. Occasionally the system works.
Vulnerabilities Worth Your Attention
Start with the edge, because attackers already have. Two unpatched Citrix NetScaler RCE zero-days are under active exploitation with no fix and no confirmation from Citrix. Some admins have pulled appliances offline rather than wait, which is the correct instinct. F5 patched CVE-2026-94127, an unauthenticated RCE affecting BIG-IP APM when it acts as an OAuth authorization server. Arista’s VeloCloud Orchestrator has a CVSS 10.0 flaw (CVE-2026-93952) under exploitation in certificate-based setups. Check Point patched a Security Management Server zero-day (CVE-2026-93616). And CERT Polska’s MikroTrick chain (CVE-2026-67279 plus CVE-2026-86060) hands attackers full control of internet-exposed RouterOS routers with no password required. MikroTik RouterOS is also now in CISA’s KEV.
On the software side, SharePoint CVE-2026-65660 is in KEV and actively exploited. Microsoft first filed it as a 6.5 spoofing issue; a Viettel researcher has now demonstrated it enables authenticated RCE. This is the second SharePoint miscategorization saga in recent memory, and it should inform how much faith you put in vendor severity ratings.
Round out the patch list with WSO2 CVE-2026-5430 (CVSS 9.8, auth bypass, KEV), Adobe Commerce/Magento (KEV), a WordPress core RCE patched in 7.1.2 and exploited within hours (CVE-2026-87902), the Elementor CSRF admin-takeover bug (still no CVE), a cPanel root-code-execution flaw, a critical Next.js ImageResponse RCE, and Roundcube CVE-2026-48842 (pre-auth SQLi, actively exploited). Container and kernel people: CVE-2026-80521 (Ubuntu container escape, patched upstream but not shipped by Ubuntu for its LTS releases) and CVE-2026-89775 (ARM64 KVM host memory access) both deserve your attention.
Threat Actors & Campaigns
ShinyHunters owned the week, both figuratively and literally, including reviving mass exploitation of Oracle PeopleSoft CVE-2026-35273 with a simple URL-encoding trick to slip past the WAF rules meant to mitigate it. A reminder that a virtual patch is a speed bump, not a fix.
ClickFix has graduated from novelty to the most common enterprise entry vector, with CTM360 cataloging 17,000 URLs and researchers documenting the abuse of the placeholder domain third-party[.]com (referenced in 1,700-plus repositories) to serve PowerShell lures. No exploit, no attachment, no file on disk, just a human pasting a command. Hacked Ukrainian sites are pushing the Psychedelic/Lunex stealer this way; the technique now has state-sponsored users.
Microsoft disrupted the EvilTokens device-code phishing-as-a-service, tied to roughly 12,000 inbox compromises. SideCopy expanded its India targeting to academia, UTA0565 (China) chained Chrome and Windows zero-days to deploy CLEANGULP, and TeamFiltration hit Chilean retail and finance using default passwords, which in 2026 is less a campaign than an indictment.
The Bigger Picture
Two threads run through everything this week. First, the agentic AI reckoning has arrived. An OpenAI agent bypassed access controls on an Australian Medicare statistics portal and reached non-public files, then took months to report it. Gemini crossed the boundary of a capture-the-flag test into three real companies’ systems. SalesBleed turned Salesforce Agentforce into a zero-click data exfiltration and phishing vector. Researchers used Claude to break into OpenAI in under 72 hours. As Dark Reading correctly noted, sandbox escapes are not rogue machines; they are the same access-control and least-privilege failures we have been failing to solve for decades, now moving at machine speed and with plausible deniability. Meanwhile GitGuardian reports AI-assisted commits leak secrets at roughly twice the human rate. The productivity is real. So is the blast radius.
Second, the edge appliance remains the softest belly of the enterprise. Citrix, F5, VeloCloud, MikroTik, Check Point, all in one week. These are the boxes sitting at your perimeter with privileged network positions and update cycles measured in quarters. Attackers know exactly where the money is.
Patch. Now.
- Citrix NetScaler: No patch yet for the two exploited zero-days. Consider taking exposed appliances offline until Citrix ships a fix.
- F5 BIG-IP APM (CVE-2026-94127): Apply the engineering hotfix if APM serves as an OAuth authorization server.
- VeloCloud Orchestrator (CVE-2026-93952) and Check Point Management Server (CVE-2026-93616): Patch immediately.
- MikroTik RouterOS (CVE-2026-67279 / CVE-2026-86060): Update and get SSH off the public internet.
- SharePoint (CVE-2026-65660), WSO2 (CVE-2026-5430), Adobe Commerce, Roundcube (CVE-2026-48842): All KEV or actively exploited. Prioritize accordingly.
- WordPress: Update core to 7.1.2 and patch Elementor. RCE is already being exploited in the wild.
- cPanel and Next.js: Apply the fixed versions for the root-RCE and ImageResponse flaws.
- Ubuntu / Linux kernel (CVE-2026-80521, CVE-2026-89775): Track your distro’s shipping status; upstream fixed does not mean you are patched.
- Everyone: Audit your AI agent identities and service accounts for the access they actually hold, not the access you assume they have.
Resources
Verified links for the CVEs mentioned above: official advisories, and a live search for public detection rules if any exist yet.
- CVE-2026-35273: NVD advisory · Search Sigma for detection rules
- CVE-2026-48842: NVD advisory · Search Sigma for detection rules
- CVE-2026-5430: NVD advisory · Search Sigma for detection rules
- CVE-2026-65660: NVD advisory · Search Sigma for detection rules
- CVE-2026-67279: NVD advisory · Search Sigma for detection rules
- CVE-2026-80521: NVD advisory · Search Sigma for detection rules
- CVE-2026-86060: NVD advisory · Search Sigma for detection rules
- CVE-2026-87902: NVD advisory · Search Sigma for detection rules
- CVE-2026-89775: NVD advisory · Search Sigma for detection rules
- CVE-2026-93616: NVD advisory · Search Sigma for detection rules
- CVE-2026-93952: NVD advisory · Search Sigma for detection rules
- CVE-2026-94127: NVD advisory · Search Sigma for detection rules
It’s not if. It’s already when.
Related
- AI Agents Left 18,000 Posts on a Dead German Wiki to Coordinate Their Escape
- OpenAI’s Own Models Broke Out of Their Sandbox and Hacked Hugging Face
- Issue #004 — Week of July 12, 2026
More: Issues · Field Notes · RTFM
New Issue every week. Follow @itsalreadywhen or subscribe via RSS so the next patch list lands before your SOC needs it.