Attackers Exploit MLflow SSRF Flaw to Steal Cloud Credentials and Secrets

Summary: Attackers are exploiting CVE-2026-64849 in MLflow, an open-source AI platform, using a flaw in its model-registry webhooks to proxy requests through affected systems and reach cloud metadata services, with watchTowr detecting indiscriminate scanning within hours of the CVE being assigned on August 17, 2026. Separately, VulnCheck reported scanning of CVE-2026-25895 in FUXA, an open-source SCADA/HMI tool, starting August 18, 2026, from a single IP address attempting to overwrite main.js with junk data via path traversal, against roughly 60 internet-exposed FUXA installations.
Key Takeaway: CISA added CVE-2026-64849 to its KEV catalog on August 19, 2026, with a September 2, 2026 patch deadline for federal agencies, and the flaw bypasses prior fixes through its handling of web redirects - so any exposed MLflow instance may already have had cloud credentials and secrets pulled from internal metadata endpoints.

Read the original at The Hacker News ↗

Topics: Vulnerabilities & Patching
Mentions: CVE-2023-33831 CVE-2026-25895 CVE-2026-25939 CVE-2026-64849
Related: ⚡ Weekly Recap: AI-Powered PLC Attacks, Git Lab Attacks, Stripe Key Leaks and More Cyber Risks ·
ZDI-26-846: MLflow Webhook Server-Side Request Forgery Vulnerability Cyber Risks · Oct 29, 2026
CVE-2026-64849: MLflow Server-Side Request Forgery Vulnerability Known Exploited Vulnerabilities · Aug 19, 2026