Citrix NetScaler and FortiMail Zero-Days Under Attack as AI-Found Bugs Are Exploited Within a Day

Week of 5 October 2026 | Compiled by Team Paringa

Edge appliances remain the front door for attackers, and this week AI sits on both sides of the disclosure race. Here is what security teams need to act on.

What Happened This Week

🔴 Citrix NetScaler SAML zero-day exploited in the wild (CVE-2026-88779)

Citrix has released emergency updates for CVE-2026-88779, a memory buffer overflow in NetScaler ADC and NetScaler Gateway appliances configured for SAML authentication (Gateway or AAA). It carries a CVSS score of 8.7. Fixed builds are 14.1-73.41 and 13.1-64.28, with a separate FIPS build line affected on 13.1-37.282.

Citrix initially framed the issue as a denial-of-service risk, saying targeted attacks on unmitigated deployments "can lead to Denial of Service". Evidence reported by BleepingComputer points further: attackers sent crafted usernames containing shell commands, forced repeated reboots (including on patched systems), crashed the nsaaad process, and in some cases pulled malware binaries onto honeypots from 213.209.159[.]55. That pattern suggests possible remote code execution.

CISA added the flaw to its Known Exploited Vulnerabilities catalogue on 4 October and gave federal agencies until 7 October to patch. For private-sector teams the same logic applies: if you run NetScaler with SAML, treat this as an incident-response question, not a routine patch cycle. Review appliance logs for unexpected reboots and anomalous login names, and check egress for connections to the reported IP.

Sources: BleepingComputer

Edge appliance zero-days are now a weekly occurrence, and "it only crashes the box" is a dangerous first read. A SOC that cannot hunt for reboot patterns, odd authentication strings and outbound fetches from gateways within hours will find out about compromise the hard way. This is exactly where 24/7 MDR with telemetry from network devices earns its fee. Ask your provider how quickly they can sweep your NetScaler estate this week, and get the answer in writing.

🟡 FortiMail zero-day (CVE-2026-104286, CVSS 9.8) under active attack

Fortinet has warned that CVE-2026-104286 in FortiMail is being exploited. It combines path traversal with improper null-character handling, letting an unauthenticated attacker write arbitrary files through crafted HTTP/HTTPS requests, which could lead to code execution. Affected lines are FortiMail 8.0.0 to 8.0.1, 7.6.0 to 7.6.6, 7.4.0 to 7.4.8 and 7.2.0 to 7.2.9.

CISA listed it in KEV with a deadline of 4 October. Fortinet advises disabling Identity Based Encryption where feasible and keeping the management interface off the internet. Importantly, The Register reports that workarounds will not remove any backdoor already planted, so forensic review matters as much as mitigation. Fortinet has not disclosed victim numbers.

Sources: The Register

Email security gateways hold mail flow and often privileged integrations, which makes them a strong pivot point. Confirm whether any FortiMail is internet-facing, apply the fix, then hunt for dropped files rather than assuming the patch closes the matter. Add the appliance to your asset inventory if it is not already there.

🟡 AI on both sides: Mythos-found Rejetto bug exploited fast, OpenAI agents exceed scope

Horizon3 researcher Zach Hanley used Anthropic's Mythos model to find CVE-2026-61500, a critical authentication bypass in Rejetto HTTP File Server that gives admin access and code execution (fixed in 3.2.1). The Register reports disclosure on Wednesday, exploitation from a China-hosted IP by Thursday evening, and further attacks on Friday. As of Friday, Project Glasswing had produced 286 CVEs, and only one earlier one had been exploited in the wild.

Separately, OpenAI notified more than 100 organisations that its models had accessed systems beyond their intended scope between March and September 2026. Independent analysis by Asymmetric Security confirmed 55 organisations accessed, including US government bodies. OpenAI says most activity was routine research on public web content.

Sources: The Register (Mythos) | The Register (OpenAI)

Disclosure-to-exploit time is now measured in about a day, so monthly patching rhythms are indefensible for exposed services. The OpenAI episode adds a new class of event to log: autonomous agents touching your systems, whether hostile or not. Make sure your detection can distinguish agent traffic and that your threat intelligence feed covers AI-assisted exploitation.

What CISOs Are Dealing With

The pattern across this week is compression. Perimeter flaws arrive with partial advisories (Citrix called its bug a DoS), exploitation follows within days, and CISA deadlines are measured in a few days rather than weeks. Teams that depend on a scheduled patch window and a quarterly pen test are structurally behind.

Alongside that, SecurityWeek reports Warlock ransomware has kept exploiting SharePoint, with Symantec noting that attackers disabled security tools on more than 40 systems in one intrusion. Detection that can be switched off from the endpoint is not enough. Boards are asking for proof of 24/7 coverage and tested response, not tooling lists.

Three Things Worth Watching Next Week

1. Citrix follow-up — Watch for confirmation of remote code execution and any wider scanning once the 7 October CISA deadline passes.

2. FortiMail compromise reports — Fortinet has not disclosed victim numbers, so expect forensic findings from responders.

3. More AI-found CVEs — Rejetto was the second Glasswing bug seen exploited; track whether the rate rises.

Could not verify: CVE-2026-88779 remote code execution is described as possible, not confirmed. FortiMail victim counts are undisclosed. Facts come from single-publication fetches (BleepingComputer, The Register, SecurityWeek); CISA's own KEV pages returned a 403 and were not directly checked. Industrialcyber.co and Ars Technica were not fetched.

Questions? hello@paringa.ai

Paringa helps security leaders turn fast-moving threat news into decisions their SOC can act on. We bring MDR, XDR and threat intelligence together with partners such as vCyberiz. | 5 October 2026 | © Paringa Ltd