# CVE-2026-64849: MLflow Server-Side Request Forgery Vulnerability

- Exploited (CISA KEV): yes, added 2026-08-19
- Published: 2026-08-17
- CVSS: 9.3 CRITICAL (v3.1)
- EPSS: 10.0% (96th percentile)
- CWE: CWE-918

MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. Starting in 3.3.0 and prior to 3.15.0, the unauthenticated POST /api/2.0/mlflow/webhooks/{id}/test endpoint calls _validate_webhook_url() in mlflow/utils/validation.py only for the original URL while mlflow/webhooks/delivery.py follows redirects and re-resolves the hostname without pinning the validated address, allowing attackers to reach internal or cloud metadata services and receive response_status and response_body. This issue is fixed in version 3.15.0.

## Edge WAF coverage (claimed by vendors, not measured)

| Vendor | Status | First rule | Days after publication |
|---|---|---|---|
| Cloudflare | No public claim | — | — |
| AWS | No public claim | — | — |
| Microsoft Azure | No public claim | — | — |
| Google Cloud | No public claim | — | — |
| Fastly | Before feed history | — | — |
| F5 | No public claim | — | — |
| Akamai | No public source | — | — |
| OWASP CRS | No public claim | — | — |

## Our test (measured)

- OWASP Core Rule Set 4.25.0, paranoia level 1, anomaly threshold 5, default CRS settings: partial (1 of 3 requests blocked), template CVE-2026-64849, 2026-11-02
- OWASP Core Rule Set 4.25.0, paranoia level 2, anomaly threshold 5, default CRS settings: partial (2 of 3 requests blocked), template CVE-2026-64849, 2026-11-02

"No public claim" means no vendor statement names this CVE; "Before feed history" means the CVE is older than the vendor feed. In both cases generic rules may still block it.

HTML: https://cve.vigilbase.com/cve/CVE-2026-64849 · JSON: https://cve.vigilbase.com/api/v1/cves/CVE-2026-64849