CISA Adds Four Exploited Flaws to KEV: Apache Axis RCE in Confluence, a Three-Day Clock, and the First LLM-Gateway Entry That Matters

CISA Adds Four Exploited Flaws to KEV: Apache Axis RCE in Confluence, a Three-Day Clock, and the First LLM-Gateway Entry That Matters

CISA Adds Four Exploited Flaws to KEV: Apache Axis RCE in Confluence, a Three-Day Clock, and the First LLM-Gateway Entry That Matters

CISA added four vulnerabilities to its Known Exploited Vulnerabilities (KEV) catalog on November 18, citing evidence of active exploitation: CVE-2026-90436 in Apache Axis (CVSS 9.8), CVE-2026-90817 in Vanderbilt University's REDCap (CVSS 9.8), CVE-2026-40217 in BerriAI LiteLLM (CVSS 8.8), and CVE-2024-10441 in Synology DiskStation Manager (CVSS 9.8). Federal agencies have been ordered to apply fixes or mitigations for the DSM, REDCap and Axis flaws by November 21, 2026 — a three-day clock — and for the LiteLLM flaw by December 2. KEV entries are not theoretical: each represents a vulnerability already observed being used against real targets.

The Apache Axis entry carries the most detailed attack narrative. Rapid7 documented exploitation of the unsafe-reflection flaw against an Atlassian Confluence Data Center instance in a customer environment, flagged on November 13: the corporate platform bundles Axis for its XML-RPC and SOAP APIs, and attackers used that embedded copy to run shell commands and scripts fetched from temporary Pinggy and Cloudflare Tunnel domains. "Just over three minutes after the first observed execution, the attackers were authenticating to MySQL with Confluence's own database credentials, querying page bodies, application links, and attachments with names associated with credentials, tokens, VPN configurations, and private keys," Rapid7 reported. Their scripts — likely AI-generated — parsed Office documents, hunted for payment card numbers using Luhn checks, and enumerated the environment. Atlassian notes Confluence is only vulnerable if the legacy "Remote API (XML-RPC & SOAP)" setting is enabled, which has been off by default since Confluence 3.0 and was fully removed in Confluence 10.0. CISA also ties the flaw to the 4EverAble2Shell campaign against N-able N-central, where it is used to deploy cryptocurrency miners.

The other three entries widen the pattern. REDCap's CVE-2026-90817 is a code injection reachable from a public survey context — exploitation requires knowledge of a valid public survey hash, but CISA describes attackers supplying a crafted file-path or stream parameter during data import processing to execute arbitrary code with no login. Synology's CVE-2024-10441, an output-encoding flaw rated 9.8, is being used in ransomware attacks against NAS devices — a two-year-old bug still paying out. And LiteLLM's CVE-2026-40217 is the one that should make every AI-infrastructure team look up: an unprotected alternate channel at the /guardrails/test_custom_code URI that allows remote code execution via bytecode rewriting, fueling LLMjacking attacks — the theft of AI model access to run other people's compute — against financial services, retail, transportation and technology organizations across the Americas and Asia-Pacific.

What Should You Do?

  1. Treat November 21 as the deadline for Axis, REDCap and Synology DSM — everyone, not just federal agencies. Map the software to your estate first: Axis hides inside products (Confluence's legacy XML-RPC/SOAP API is the documented case), REDCap runs in research institutions, DSM on NAS gear.
  2. Disable the legacy Confluence Remote API (XML-RPC & SOAP) if it's still enabled on your instance — Confluence 10.0 removed it entirely, so the real question is whether an old version is still running.
  3. Audit LiteLLM and every AI gateway's test/debug endpoints. /guardrails/test_custom_code is exactly the kind of convenience endpoint that ships enabled; if you run LLM gateways, expose them only on internal networks and rotate every provider API key.
  4. Check REDCap public surveys and Synology exposure: restrict who can obtain public survey hashes, and take Synology DSM instances off the open Internet — ransomware crews are already in the KEV-confirmed door.

The WAF Angle

Three of these four flaws are HTTP-reachable web endpoints, which makes them exactly what a WAF is for — and in each case the tell is structural. The Axis exploit reaches a SOAP API with unexpected payloads; the REDCap flaw supplies a crafted file-path or stream parameter to a data-import route; LiteLLM's bytecode-rewriting URI is a known, fixed path that should never see traffic from outside. A virtual-patch rule blocking requests to /guardrails/test_custom_code on your AI gateway, or anomalous parameters on import endpoints, buys the time that CISA's three-day clock does not. The Synology lesson echoes earlier coverage of Zyxel's GS1900 patch-gap: internet-exposed management planes on edge devices get found and exploited no matter how boring they look — put a real filter or a VPN in front of them, not a forwarded port.

Sources