LiveThreat Vulnerability Brief — CVE‑2026‑2635
What It Is — MLflow (an open‑source machine‑learning lifecycle platform) ships with a hard‑coded default password in its basic_auth.ini file. This flaw enables remote attackers to bypass authentication entirely and execute arbitrary code with administrator privileges. Exploitability — No authentication is required; a remote attacker can directly exploit the default credentials. CVSS v3.1 base score 9.8 (Critical).
Affected Products — Vendor: MLflow; Product: MLflow (all versions prior to the 2.12.0 patch).
TPRM Impact — Organizations that embed MLflow in internal pipelines, SaaS offerings, or third‑party data‑science services face a supply‑chain risk: an attacker could gain admin access, tamper with model artifacts, exfiltrate proprietary data, or pivot to downstream systems.
Recommended Actions — 1️⃣ Deploy the MLflow update that removes the default credentials (see GitHub pull #19260). 2️⃣ Audit any existing basic_auth.ini files for leftover defaults and replace them with strong, unique passwords or integrate with external identity providers. 3️⃣ Enforce network‑level segmentation and monitor for anomalous API calls to MLflow endpoints. 4️⃣ Conduct a rapid risk assessment of all downstream services that consume MLflow‑managed models. Source: Zero Day Initiative Advisory ZDI‑26‑111