// Security Advisory Triage
MLflow path traversal allows arbitrary file writes via trace IDs
GitHub advisories — pip · published 2026-10-24 16:01 UTC
WHAT CHANGEDGitHub published CVE-2026-53995 on 20 October 2026, a high-severity path traversal vulnerability in mlflow's FileStore trace ID handling. The flaw allows arbitrary file writes. Patched in mlflow 3.17.2.
WHY IT MATTERSAn attacker who can control trace IDs can write files anywhere the mlflow process has permissions, potentially overwriting application code, configuration, or credentials. This affects any deployment using mlflow's FileStore backend for trace logging.
WHAT TO DOUpgrade to mlflow 3.17.2 or later. If you use mlflow for experiment tracking or model registry, check whether FileStore is your trace backend and prioritize the update if so.
Model-reported confidence 85% (not measured accuracy)
Primary source ↗
written by anthropic/claude-sonnet-4.5
Show the reasoning — raw diff, triage verdict, confidence
RAW DIFF
+ghsa_id: GHSA-wqmh-qccw-pqwf +summary: MLflow FileStore trace IDs allow path traversal and arbitrary file writes +severity: high +published_at: 2026-10-20T00:31:59Z +cve_id: CVE-2026-53995 +vulnerabilities[].package.name: mlflow +vulnerabilities[].first_patched_version: 3.17.2
HOW IT WAS JUDGED
Triage signals below are model-reported assertions, preserved verbatim; they are not independent arithmetic verification.
| triage verdict | MATERIAL at 0.92, needed 0.5 |
|---|---|
| triage reasoning | MLflow is widely deployed in ML pipelines and data infrastructure. Path traversal + arbitrary file writes is exploitable and patched. |
| signals | new advisory published · high severity rating · path traversal and arbitrary file write vulnerability · affects mlflow package · patched version available (3.17.2) · mlflow is a common ML ops dependency in data/ML infrastructure |
| suppression | — |
| triage model | anthropic/claude-haiku-4.5 ($0.00000) |
| writer model | anthropic/claude-sonnet-4.5 ($0.00000) |