CISA added four vulnerabilities to its Known Exploited Vulnerabilities (KEV) catalog on November 18, meaning it has evidence they are being used in attacks. According to SecurityOnline's breakdown of the entries, they are a code-execution flaw in the LiteLLM AI gateway, an unauthenticated remote code execution bug in Vanderbilt's REDCap research-data platform, a 2024 Synology DSM flaw, and the Apache Axis 1.x bug we covered on November 17.
Why it matters: two of these sit on unusually sensitive data. LiteLLM proxies hold the API keys for every model provider an organisation uses. REDCap often stores patient and human-subjects data at hospitals, universities and public-health agencies, Dataminr says. SecurityOnline reports federal agencies get just three days (to November 21) for the REDCap, Synology and Axis entries, and until December 2 for LiteLLM. We could not open CISA's own alert page to confirm the dates, so treat them as SecurityOnline's reading.
The four entries
| CVE | Product | What it is | Fixed in | Federal deadline (per SecurityOnline) |
|---|---|---|---|---|
| CVE-2026-40217 | BerriAI LiteLLM | Sandbox escape in the custom-code guardrail tester; needs a valid key | 1.83.10; full related chain fixed in 1.83.14-stable | Dec 2 |
| CVE-2026-90817 | Vanderbilt REDCap 13.3.0+ | Unauthenticated RCE via a public survey link and the Data Import route | 16.0.49, 17.3.10, 17.4.4 | Nov 21 |
| CVE-2024-10441 | Synology DSM / BeeStation OS | Unauthenticated code execution in a system plugin daemon | DSM 7.2.2-72806-1 and branch equivalents; BSM 1.1-65374 | Nov 21 |
| CVE-2026-90436 | Apache Axis 1.x (end of life) | Unsafe reflection in SOAP fault parsing, before authentication | No release; upstream commit only | Nov 21 |
SecurityOnline lists all four at CVSS 9.8 except LiteLLM, which Red Hat also scores 8.8.
LiteLLM: probed since the day of the fix
The LiteLLM bug lets someone who holds an API key send Python to the /guardrails/test_custom_code endpoint and break out of its sandbox, Red Hat's entry says. Zenity Labs reported in June that its honeypots logged 569 POST requests to that endpoint between April 15 and May 24, starting on the day the fix shipped. The requests tried 47 guessed master keys such as sk-1234, and later payloads tried to read provider keys from environment variables. Zenity notes the default Docker image runs as root, so an escape gives root.
Obsidian Security showed the flaw can sit at the end of a chain. Two other bugs, CVE-2026-47101 and CVE-2026-47102, let a low-privilege user promote themselves to admin. Obsidian says 1.83.14-stable (April 25) is the first release that closes the whole chain. Version advice varies by source: SecurityOnline points to "builds after April 8", and Zenity names 1.83.10. Separately, Forkast reported in September that CISA had already listed an older LiteLLM bug, CVE-2026-59822, which is fixed only in 1.84.0. So the practical target is 1.84.0 or later.
REDCap: no login needed
Dataminr says the REDCap flaw was disclosed on September 20. An attacker who has a valid public survey hash, which is the token in any public survey link, can route requests to the Data Import controller and run code as the web server user. Dataminr notes a public proof of concept exists. Dataminr warns that some REDCap instances run outside central IT, which makes unpatched copies easy to miss.
What to do
- LiteLLM: upgrade to 1.84.0 or later. Replace any default or guessable master key. Keep the admin and control-plane routes off the internet, and alert on outside POSTs to
/guardrails/test_custom_code. If you were exposed, rotate every provider key the proxy holds. Obsidian also advises checkingproxy_adminaccounts and the callbacks inconfig.yaml, which the admin UI does not show. Red Hat customers should apply the errata SecurityOnline cites for Ansible Automation Platform 2.6 and OpenShift AI 3.3. - REDCap: move to 16.0.49 LTS, 17.3.10 LTS or 17.4.4. Find every instance at 13.3.0 or later, including departmental ones. Dataminr also advises restricting public survey links and watching for unusual requests to
__passthruand Data Import routes. - Synology: update DSM, BeeStation or DSMUC to the fixed builds. This is the same December 2024 fix tied to C3RB3R ransomware attacks earlier this month.
- Apache Axis: follow our Confluence and N-central guide.
Why three days
Under CISA's Binding Operational Directive 26-04, RapidFort explains, a KEV bug that gives an attacker total control gets a three-day deadline plus forensic triage. That means checking whether the system was already breached, not just patching it. Patching quickly does not show that you were not hit. SecurityOnline says CISA ordered triage for the Synology, Axis and REDCap entries. CISA lists ransomware use as unknown for all four.
The directive binds only US federal civilian agencies. In practice, the KEV catalog is the patch list most security teams follow.
Sources
- SecurityOnline — CISA Adds 4 Exploited Flaws to KEV: Synology DSM, LiteLLM securityonline.info
- Parker Nathan — #Cybersecurity News: November 18, 2026 theparkernathan.substack.com
- Zenity Labs — Threat Actors Are Trying to use LiteLLM's Guardrail Tester to Run Code as Root labs.zenity.io
- Obsidian Security — Breaking LiteLLM: From Low-Privilege User to Admin and RCE obsidiansecurity.com
- Red Hat — CVE-2026-40217 access.redhat.com
- Dataminr — Unauthenticated RCE Hits REDCap: CVE-2026-90817 Explained dataminr.com
- RapidFort — BOD 26-04 explained: how CISA now sets KEV deadlines rapidfort.com
- Forkast — Three-of-Seven CISA KEV Additions Now Target AI Infrastructure forkast.news