
CISA Expands Known Exploited Vulnerabilities Catalog
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) added four actively exploited software vulnerabilities to its Known Exploited Vulnerabilities (KEV) catalog on November 18, 2026, citing active in-the-wild abuse across AI middleware, legacy application libraries, academic data platforms, and network storage appliances. As reported by The Hacker News, the advisories cover flaws in BerriAI LiteLLM, Apache Axis, Vanderbilt University Research Electronic Data Capture (REDCap), and Synology DiskStation Manager (DSM).
Threat intelligence disclosures linked to the four bugs show attackers chaining public-facing exploits into credential theft, unauthorized AI model access, cryptocurrency mining, and ransomware deployment. Following the KEV additions, CISA directed federal civilian executive branch agencies to remediate or mitigate the Synology DSM, REDCap, and Apache Axis vulnerabilities by November 21, 2026, and set a December 2, 2026 deadline for addressing the LiteLLM flaw.
Technical Architecture and Vulnerability Mechanics
Three of the four cataloged flaws carry CVSS severity scores of 9.8, while the fourth carries a score of 8.8. In AI infrastructure software, CVE-2026-40217 (CVSS 8.8) is an unprotected alternate channel vulnerability in BerriAI LiteLLM that enables remote arbitrary code execution via bytecode rewriting at the /guardrails/test_custom_code URI. Because LiteLLM serves as a proxy and routing layer for large language model APIs, arbitrary code execution inside that service boundary exposes underlying inference credentials and runtime configurations.
The remaining three vulnerabilities affect foundational enterprise and research components. CVE-2026-90436 (CVSS 9.8) is an unsafe reflection flaw in Apache Axis 1.x that lets an unauthenticated remote attacker execute arbitrary code through XML-RPC and SOAP interfaces. CVE-2026-90817 (CVSS 9.8) is a code injection vulnerability in REDCap triggered when an unauthenticated attacker sends crafted HTTP requests to an unintended controller route from a public survey context, passing a manipulated file-path or stream parameter during data import processing while supplying a valid public survey hash. Rounding out the additions, CVE-2024-10441 (CVSS 9.8) stems from improper encoding or escaping of output in Synology DSM, allowing remote arbitrary code execution via unspecified vectors.
LLMjacking Escalation and Multi-Stage Enterprise Intrusions
Active exploitation of CVE-2026-40217 was documented this week by the Google Threat Intelligence Group, which observed threat actors abusing LiteLLM instances to escalate limited-access inference keys into full administrative credentials and establish persistence. Those intrusions fueled LLMjacking campaigns targeting financial services, retail, transportation, and technology organizations across North America, South America, and the Asia Pacific region, converting enterprise AI gateways into hijacked compute and API assets.
Enterprise collaboration servers running bundled Axis libraries have faced similarly rapid post-exploitation. Rapid7 investigated an incident flagged on November 13, 2026, in which attackers exploited CVE-2026-90436 against Atlassian Confluence Data Center deployments that rely on Axis for XML-RPC and SOAP APIs. Operators fetched shell scripts from temporary Pinggy and Cloudflare Tunnel domains and, within just over three minutes of initial execution, authenticated to MySQL using Confluence database credentials to query page bodies, application links, and attachments for tokens, VPN configurations, and private keys. Rapid7 noted that the attackers deployed likely AI-generated scripts against /var/atlassian/application-data/confluence/attachments/v4/ to parse Office documents and scan for payment card numbers filtered with Luhn checks, then queried the internal Atlassian Crowd API, attempted Kerberos AS-REP Roasting, pulled data from a production cloud data warehouse with a captured service account, and logged into Microsoft 365 roughly 12 hours later using legacy authentication from IP address 192.238.253[.]176. Separately, CVE-2026-90436 has been chained with 4EverAble2Shell flaws in N-able N-central to deploy cryptocurrency miners.
Resource hijacking and extortion have likewise hit research and storage systems. Safe Security reported late last month that an unknown threat actor exploited REDCap flaw CVE-2026-90817 to stage a transient web shell, retrieve an XMRig miner disguised as image files via an in-memory loader, masquerade the process as syslog-ng, and maintain persistence through cron jobs and an added SSH key. Meanwhile, Synology disclosed earlier in November that CVE-2024-10441 is being actively exploited against Network Attached Storage devices to deploy C3RB3R ransomware.
Operational Impact on AI Teams and Infrastructure Workflows
For engineering teams operating AI gateways and enterprise collaboration stacks, the combined KEV warnings underscore how guardrail testing endpoints and legacy protocol bridges collapse trust boundaries when exposed to untrusted networks. In LiteLLM environments, organizations that issue scoped inference keys to developers or internal services cannot treat key segmentation as sufficient containment if the /guardrails/test_custom_code endpoint remains accessible for bytecode manipulation.
At the same time, security teams managing Confluence, REDCap, and Synology appliances face compressed response windows against automated enumeration. When attackers move from an initial Axis reflection exploit to internal database extraction and cloud data warehouse queries in minutes, static perimeter monitoring leaves little margin for manual triage, forcing administrators to audit active API modules, restrict public survey import routes, and isolate NAS management interfaces.
End-of-Life Constraints and Regulatory Deadlines
Remediation paths vary sharply across the four affected products because of software lifecycle constraints. Apache Axis 1.x has reached end-of-life status and remains without an official vendor patch for CVE-2026-90436, leaving organizations to migrate to an alternative SOAP engine such as Apache Axis 2/Java or compile a custom Axis build incorporating a July 12, 2026 commit fix. Within Atlassian Confluence, the vulnerable Remote API (XML-RPC and SOAP) setting was disabled by default for new installations starting in Confluence 3.0 and completely removed in Confluence 10.0, limiting exposure to legacy or explicitly re-enabled configurations.
Other caveats govern risk prioritization ahead of CISA's November 21 and December 2 federal compliance cutoffs. Successful exploitation of REDCap's CVE-2026-90817 requires an attacker to know a valid public survey hash, whereas Synology has not detailed the exact input vectors behind CVE-2024-10441 despite confirming active ransomware deployment. Organizations running any of the four products must verify patch levels, disable unnecessary legacy or guardrail execution endpoints, and rotate exposed service credentials immediately.
Automated Vertex AI (gemini-4-hydrogen) primary-source report; reviewed 2026-11-19.



