On November 18, the Cybersecurity and Infrastructure Security Agency added three vulnerabilities to its Known Exploited Vulnerabilities (KEV) catalog. KEV is CISA's list of bugs it has evidence are being used in real attacks. The three products have little in common. One is a research-data platform from Vanderbilt University. One is a SOAP web-services library that is end-of-life. One is a gateway that sits in front of large language models.
Two of the entries come with a three-day remediation deadline, November 21. The third is due December 2. The KEV entries cite Binding Operational Directive 26-04, which governs how quickly federal agencies must act on KEV entries. As usual, the catalog does not say who is exploiting these flaws or how widely, and it lists known ransomware use as "Unknown" for all three.
The three entries at a glance
- CVE-2026-90817, Vanderbilt REDCap. Unauthenticated code injection (CWE-73/94). Due November 21. Forensic triage flagged "Yes."
- CVE-2026-90436, Apache Axis 1.x. Unauthenticated unsafe reflection (CWE-470). Due November 21. CISA says the product could be end-of-life and advises discontinuing use. Forensic triage flagged "Yes."
- CVE-2026-40217, BerriAI LiteLLM. Unprotected alternate channel (CWE-420) leading to remote code execution. Due December 2. Forensic triage flagged "No." Known ransomware use is listed as "Unknown."
The forensic triage flag on the first two is the one to notice. It signals that patching may not be enough, and that affected organizations should also check whether they have already been compromised. Applying a fix three days after a bug lands in KEV does not undo anything an attacker did beforehand.
REDCap: a survey link as the way in
REDCap (Research Electronic Data Capture) is a web application from Vanderbilt University for building research surveys and databases.
According to the NVD record, published September 20, the flaw is an unauthenticated remote code execution bug in REDCap's survey passthrough routing and Data Import processing. An attacker manipulates HTTP requests from a public survey context to reach a controller route that was never meant to be reachable from there, and supplies a crafted file-path or stream parameter. The result is code running on the server.
There is one precondition. The attacker needs a valid public survey hash, the identifier that appears in a REDCap public survey link. That is less of a barrier than it sounds. Public surveys are built to be shared, so their links get emailed, posted, and embedded in recruitment pages. Any instance with even one public survey online should be treated as exposed.
NVD lists the vulnerability as existing in REDCap 13.3.0 and higher, with a CVSS 3.1 score of 9.8 (Critical) supplied by Securifera as a secondary source. The NVD record marks versions 16.0.49, 17.3.10 and 17.4.4 as unaffected, which points to those builds as the fixed releases. Confirm the right build for your release line with Vanderbilt and your REDCap support channels.
The bug was published in mid-September and reached KEV in mid-November. That two-month gap shows why "critical, unauthenticated, internet-facing" deserves priority even before anyone confirms exploitation.
Apache Axis 1.x: a critical bug that will never get a release
Apache Axis 1.x is an old Java framework for SOAP web services. NVD marks this CVE "UNSUPPORTED WHEN ASSIGNED," which means the product was already out of support when the bug was disclosed.
The Apache project's advisory, posted to the oss-security mailing list on November 6, rates the issue critical and says it affects every Axis 1.x release through 1.4. The problem is the order of operations. When Axis processes a SOAP Fault carrying an exceptionName element in its detail, it loads the named class and instantiates it before checking that the class is actually an AxisFault. If a suitable class is present on the application's classpath, that can lead to remote code execution. This is the "unsafe reflection" in CWE-470: code chooses which class to create based on input an attacker controls.
The advisory makes two points that matter most for defenders. First, the fault is parsed before any per-service handler chain runs, so authentication handlers configured on a service do not protect against this. Second, the project does not expect to release a fixed Axis 1.x version. It recommends migrating to a different SOAP engine, such as Apache Axis 2/Java, or using a build of Axis with the patch from commit 41f3b4e8 applied. CISA's KEV entry advises users to discontinue use or transition to a supported version.
The flaw was credited to José Apari-Pantigozo of OKIOK. It went from public advisory to KEV listing in 12 days.
LiteLLM: a test endpoint that runs code
LiteLLM from BerriAI is a proxy and SDK that gives applications one interface to many LLM providers. Organizations often deploy such a proxy as a central gateway holding API keys for several AI services, which makes it a valuable target.
The KEV entry describes CVE-2026-40217 as an unprotected alternate channel (CWE-420) that could allow remote attackers to execute arbitrary code through bytecode rewriting at the /guardrails/test_custom_code URI. CWE-420 describes a secondary path into functionality that lacks the protections applied to the main one. In plain terms, a route that appears intended for testing custom guardrail code turned into a way to run arbitrary code.
The sources reviewed for this article do not specify affected or fixed LiteLLM versions. CISA lists ransomware use as "Unknown," and the December 2 deadline gives agencies about two weeks rather than three days.
What It Means for Defenders
The KEV deadlines formally bind federal civilian agencies, but the catalog is a useful free signal on what attackers are actually using. Contractors, including those working toward CMMC, should consider treating these dates as their own.
REDCap (CVE-2026-90817), due November 21:
- Inventory every REDCap instance, including department-run and test servers. NVD lists 13.3.0 and higher as affected.
- Upgrade to a build NVD lists as unaffected (16.0.49, 17.3.10 or 17.4.4) or a later release, and confirm the correct build for your line with Vanderbilt and your REDCap support channel.
- Until it is patched, consider disabling public surveys or restricting the instance to trusted networks, since a valid public survey hash is the only precondition.
- Because CISA flags forensic triage, look over web and application logs for unusual requests coming from public survey contexts, unexpected Data Import activity, and new or modified files on the server. If you find signs of compromise, rotate the credentials the REDCap server holds, including database and integration accounts.
Apache Axis 1.x (CVE-2026-90436), due November 21:
- Search your software for Axis 1.x JARs, including inside vendor products and older internal applications. Every version through 1.4 is affected.
- No fixed release is planned. The options are migrating (for example to Axis 2/Java), applying the Apache patch commit 41f3b4e8, or taking the service offline. CISA's guidance is to discontinue use.
- Do not count on per-service authentication handlers. The advisory says they do not block this flaw.
- If the application comes from a vendor, ask them directly whether it bundles Axis 1.x and what their remediation plan is.
- Do a forensic review of any internet-exposed Axis endpoint, as the triage flag indicates.
LiteLLM (CVE-2026-40217), due December 2:
- Find every LiteLLM proxy deployment, including ones data science or developer teams stood up on their own.
- Apply the mitigations in BerriAI's advisory and release notes, since the sources here do not name a fixed version.
- Block external access to the
/guardrails/test_custom_codeendpoint and keep LiteLLM administrative interfaces off the public internet. - A compromised LLM gateway can expose every provider key stored on it. If a deployment was reachable and unpatched, consider rotating those API keys and reviewing provider usage for unexpected spend or activity.
Across all three, CISA has published no attribution and no scale of exploitation. The absence of that detail is not reassurance. Unauthenticated, internet-reachable code-execution flaws with three-day deadlines show that CISA considers them urgent.
Sources
- CISA Known Exploited Vulnerabilities Catalog, JSON feed (entries added 2026-11-18)
- NVD record for CVE-2026-90817, REDCap (read via NVD API)
- oss-security: CVE-2026-90436, Apache Axis 1.x (EOL) unauthenticated remote class instantiation via SOAP fault exceptionName (Apache project advisory)
- NVD record for CVE-2026-90436, Apache Axis (read via NVD API)