CVE-2026-111058: Unauthenticated RCE in mljar Mercury via /api/kernels
What's Going On
CVE-2026-111058 is a missing-authorization flaw (CWE-862) in mljar Mercury, a Jupyter-notebook-to-webapp framework. Affected versions expose the Jupyter-style /api/kernels endpoint without any authentication check. Because kernel endpoints are designed to accept arbitrary code for execution, an unauthenticated attacker can simply create a kernel and submit code over its WebSocket/channels interface, resulting in direct remote code execution on the host running the Mercury app server.
Affected Software
- Vendor: pip (PyPI package)
- Product: Mercury (mljar Mercury)
- Affected versions: <= 3.2.5
- No patched version is currently published.
Exploitation Status
CVE-2026-111058 carries a CVSS score of 9.8 (critical) and is classified as PoC public: a working proof of concept exists in the wild, but it has not been confirmed as weaponized in active campaigns or added to CISA's KEV catalog at this time. Given the triviality of exploitation — no authentication, no privilege, no user interaction required — defenders should treat any internet-facing or otherwise reachable Mercury instance running an affected version as an immediate priority for patch tracking and compensating controls.
How Our Detection Catches It
Our detection targets the exploitation chain at two layers:
- Network/application layer: unauthenticated HTTP requests to
/api/kernels, including POST requests that create new kernels and GET requests used to enumerate existing ones, with particular attention to rapid, repeated kernel-create activity consistent with automated exploitation. - Host/process layer: suspicious child processes spawned by the Mercury/Python server process, which is the expected signature of arbitrary code executing inside a compromised kernel.
This logic is shipped across all seven of our SIEM dialects — Microsoft Sentinel (KQL), Splunk (SPL), Elastic (EQL), QRadar (AQL), Sumo Logic, Chronicle (YARA-L), and CrowdStrike (CQL) — so the same exploitation pattern can be detected regardless of which platform ingests your web and endpoint telemetry.
Full Detection
For the complete, ready-to-deploy queries, investigation guidance, and atomic tests for this detection, see the CVE-2026-111058 detection page.