CVE-2026-64849
MLflow: Unauthenticated full-read SSRF in webhook delivery: _validate_webhook_url bypassed via unvalidated HTTP redirects (and DNS rebinding)
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:L/A:N
Description
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.
Official description from the CVE record, published by GitHub_M. Copyright © 1999-2026 The MITRE Corporation. See /security/cve/legal for the CVE license. View the record on cve.org
Key data
CVSS 3.1 Critical 9.3
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:L/A:N
Attack vector: Network
The vulnerable component can be reached over a network, possibly from the Internet.
Attack complexity: Low
No special conditions are needed. The attack can be repeated at will.
Privileges required: None
No account or privileges are needed.
User interaction: None
No action from any user is needed.
Scope: Changed
The impact can reach other components beyond the vulnerable one.
Confidentiality impact: High
Serious or total effect on the secrecy of data.
Integrity impact: Low
Limited effect on the trustworthiness of data.
Availability impact: None
No effect on availability.
| Metric | Value | What it means |
|---|---|---|
| Attack vector | Network | The vulnerable component can be reached over a network, possibly from the Internet. |
| Attack complexity | Low | No special conditions are needed. The attack can be repeated at will. |
| Privileges required | None | No account or privileges are needed. |
| User interaction | None | No action from any user is needed. |
| Scope | Changed | The impact can reach other components beyond the vulnerable one. |
| Confidentiality impact | High | Serious or total effect on the secrecy of data. |
| Integrity impact | Low | Limited effect on the trustworthiness of data. |
| Availability impact | None | No effect on availability. |
CVSS is owned and managed by FIRST.Org, Inc. (FIRST) and is used by permission. Each CVSS score on this site is shown together with its vector string, as published in the CVE record by the source named next to it. Learn more at first.org/cvss.
EPSS
- Score
- 0.10017 (10.0% probability)
- Percentile
- 95.7%
- Scores of
- Nov 10, 2026
- Model
- v2026.06.15
CISA Known Exploited Vulnerabilities
- Name in the catalog
- MLflow Server-Side Request Forgery Vulnerability
- Added to the catalog
- Aug 19, 2026
- Federal due date
- Sep 2, 2026
CISA assessment as of Aug 20, 2026
- Exploitation
- Recorded by CISA as active
- Automatable
- yes
- Technical impact
- partial
Weakness (CWE)
- CWE-918Server-Side Request Forgery
Affected products
mlflow
- affected: >= 3.3.0, < 3.15.0
References
Timeline
- ID reserved
- Record published
- Added to CISA KEV
- Record last updated
Look up another CVE
About CubePath
Data from the CVE Program (MITRE), CISA KEV and FIRST EPSS. CVSS is owned by FIRST and used by permission. Not affiliated with or endorsed by MITRE, CISA or FIRST. Data as of Nov 10, 2026, 19:09 UTC. Sources and licenses