CVE-2026-64849MLflow Server-Side Request Forgery Vulnerability
CVE-2026-64849 is a server-side request forgery in MLflow, rated critical (9.3 on CVSS 3.1).
Why this priority: Exploited in the wild: listed in CISA's KEV catalog.
Published · Updated · NVD status: Analyzed
What it is
The server can be made to fetch an address of the attacker's choosing, including internal services and cloud metadata endpoints.
How it is reached
An attacker can exploit it over the network, without an account, and nobody has to click anything.
What an attacker gains
A successful attack can read the data it protects. In practice, this kind of flaw lets an attacker make the server send requests into internal networks or cloud metadata. The damage can spread past the vulnerable component into others.
Is it being exploited?
CISA added it to the Known Exploited Vulnerabilities catalog on 2026-08-19: it is being exploited in the wild. It can be exploited automatically, at scale. EPSS estimates a 10% chance of exploitation activity in the next 30 days, higher than 96% of scored CVEs.
What to do
- Treat it as urgent: it is on CISA's KEV list. CISA's deadline for US federal agencies: 2026-09-02.
- Upgrade MLflow to 3.15.0 or later. The patch notes are linked below.
- Until it is patched, keep it off the internet or behind a VPN, allowlist or authenticating proxy.
- Check exposed instances for signs of compromise, not only for the version.
Description from NVD
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.
What this means for your architecture.
A CVE names a product. Whether it matters depends on where that product sits in your systems: what reaches it, which trust boundary it guards and what it can reach once compromised. Veriom maps your components, their relationships and your trust boundaries from your code, cloud and delivery evidence, so you can see where a component like this sits. The radar itself does not look at your systems.
Components it touches
- Services that fetch URLs or call other hosts for a caller. The flaw sits between the application and the internal network behind it.
Trust boundaries in play
- Internet-facing entry points. No account is needed. Any instance reachable from outside your perimeter, through a load balancer, gateway, VPN or exposed admin port, is in scope.
- Isolation between components. The impact crosses out of the vulnerable component into others it should be isolated from.
- Between the application and the internal network behind it. An attacker who crosses it can make the server send requests into internal networks or cloud metadata.
Questions to answer about your own systems
- Where does MLflow run in your systems, and at which version?
- Which of those instances can be reached from the internet or partner networks?
- What can it reach once compromised: which data stores, secrets and services?
Read more
Patches and commits
Exploits and proofs of concept
Sources
- CVE records and CVSS scores from the NVD CVE API. This product uses the NVD API but is not endorsed or certified by the NVD.
- Exploitation status from CISA’s Known Exploited Vulnerabilities catalog.
- Exploit likelihood from FIRST’s Exploit Prediction Scoring System (EPSS), refreshed daily.
- Explanations are written by Veriom from fixed templates over each record’s data, not by a person reviewing each CVE. Descriptions are quoted from the records as published.