Active Exploitation of MLflow SSRF and FUXA RCE Flaws Exposes Cloud Credentials
Written by AI from the linked story. Not reviewed by an editor. How this works
Attackers are actively exploiting two critical vulnerabilities — an SSRF flaw in MLflow (CVE-2026-64849) and a remote code execution bug in FUXA (CVE-2026-25895) — almost immediately after public disclosure, highlighting the dangerously narrow window organizations have to patch before threat actors begin scanning. The MLflow SSRF vulnerability is particularly severe because it allows attackers to pivot from a web-facing service to internal cloud metadata endpoints, harvesting credentials that can grant broad access to cloud infrastructure. The FUXA flaw compounds the risk by enabling full remote code execution through file system manipulation, potentially giving attackers persistent footholds in operational technology environments. These incidents underscore a recurring pattern: unpatched internet-facing systems in AI/ML pipelines and industrial control contexts are high-value, often under-secured targets.
Tactical Insight
Immediate actions
- Apply vendor-supplied patches or upgrade MLflow and FUXA to the latest secure versions without delay.
- Block or restrict outbound requests from MLflow and similar ML platform services to cloud metadata endpoints (e.g., 169.254.169.254) using firewall rules or egress filtering.
- Audit all internet-facing ML and OT/SCADA services to confirm exposure scope and isolate any compromised systems.
Long-term improvements
- Implement an automated vulnerability scanning and patch management program that prioritizes internet-facing and cloud-connected assets.
- Enforce least-privilege IAM policies so that even if cloud credentials are stolen via SSRF, their blast radius is minimized.
- Maintain a current, accurate software inventory (SBOM) covering ML frameworks and industrial control platforms to accelerate response when new CVEs are disclosed.
Detection measures
- Deploy runtime monitoring to alert on unusual outbound HTTP requests from ML services targeting internal or metadata IP ranges.
- Configure SIEM rules to detect rapid exploitation indicators such as scanning patterns or anomalous file system access associated with FUXA exploitation.
- Enable cloud provider audit logging (e.g., AWS CloudTrail, Azure Monitor) to detect credential usage from unexpected sources or geolocations.