MLflow flaw exploited to steal cloud credentials
Attackers are exploiting a new server-side request forgery (SSRF) flaw in MLflow to steal cloud metadata and secrets, watchTowr reported on August 17, 2026.
Source: The Hacker News · August 18, 2026 at 11:01 PM · AI-assisted report
Single-source
KUALA LUMPUR, 19 AUGUST 2026 —
Listen to this article
DomainFork Audio · read aloud
Attackers are exploiting a new server-side request forgery (SSRF) flaw in MLflow to steal cloud metadata and secrets, watchTowr reported on August 17, 2026.
Market Impact
The flaw, tracked as CVE-2026-64849, lets an attacker proxy requests through exposed MLflow instances and reach internal cloud services, according to watchTowr principal threat intelligence specialist Yordan Ganchev. The cybersecurity firm detected indiscriminate scanning for exposed MLflow systems within hours of the CVE being assigned. Evidence from global honeypot telemetry shows attackers targeting cloud-hosted MLflow deployments to extract credentials from well-known internal IP addresses and services.
Ganchev said the bug bypasses prior fixes because of how it handles web redirects.
Affected organizations should patch exposed systems, review audit logs for signs of compromise, and check whether sensitive credentials have been exposed.
Separately, attackers are scanning for a path traversal flaw, CVE-2026-25895, in FUXA, an open-source SCADA/HMI platform. VulnCheck vice president of research Caitlin Condon said a single IP address has been broadly scanning the internet for vulnerable FUXA instances since August 18, 2026. About 60 FUXA installations are exposed to the public internet. Condon said attackers are attempting to overwrite main.js via the path traversal flaw, but no remote code execution payloads have been observed.
FUXA has seen two other vulnerabilities—CVE-2026-25939 and CVE-2023-33831—actively exploited in the past year, with the latter targeted as recently as the day before.