CVE-2026-4035: MLflow AI Gateway Environment Variable Credential Exfiltration
- Vendor
- Lfprojects
CVE-2026-4035 affects MLflow versions before 3.11.0 with AI Gateway APIs enabled. AI Gateway secret handling resolved an api_key value prefixed with "$" as a reference to an environment variable in the MLflow server process. The resolved value was subsequently included in provider authentication headers for requests sent to the user-configured upstream API base. An attacker can create a gateway secret referencing a sensitive server-side environment variable, configure an attacker-controlled upstream endpoint, and invoke a gateway request to receive the resolved secret. MLflow 3.11.0 changes this behavior by making environment-based API-key resolution conditional on the MLFLOW_GATEWAY_RESOLVE_API_KEY_FROM_ENV setting, which defaults to disabled for API-driven gateway deployments.
Impact, mitigation & remediation
What it means and what to do now: patch path, interim mitigations, and the assume-compromise checklist.
Impact
What an attacker gets, and what they’ve been doing with it.
Attackers can exfiltrate arbitrary environment variables available to the MLflow server, including cloud credentials used to access model-artifact storage. Compromised artifact-storage credentials may permit replacement or poisoning of model artifacts. Where downstream consumers deserialize poisoned Python model artifacts, this can lead to arbitrary code execution in those downstream environments. The reported CVSS v3.1 vector is AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:L/A:L; default deployments without authentication may permit unauthenticated exploitation.
Mitigation
If you can’t patch tonight, do this now.
Until remediation is complete, disable AI Gateway functionality where feasible; restrict gateway-secret creation and gateway invocation to trusted administrators; enable and enforce authentication rather than exposing a default unauthenticated MLflow deployment; and apply outbound egress controls so the MLflow service cannot reach untrusted upstream endpoints. Restrict the MLflow process environment to only required secrets and monitor gateway configuration changes and outbound authentication requests.
Remediation
Patch, then assume compromise.
Upgrade MLflow to version 3.11.0 or later. Ensure MLFLOW_GATEWAY_RESOLVE_API_KEY_FROM_ENV is set to False. This setting is automatically enabled for backward compatibility in legacy YAML gateway deployments launched with the legacy gateway command, so operators of such deployments must explicitly disable it after upgrading. Rotate potentially exposed server and cloud credentials, review gateway secrets for environment-variable API-key references, and audit model artifacts and artifact-storage access for unauthorized modification.
Exploits
No public exploits tracked yet. Mallory keeps watching.
Affected products & vendors
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
References
Frequently asked questions
- Is CVE-2026-4035 exploited in the wild?
- Yes. As of Sep 25, 2026, Mallory has reports of exploitation in the wild.
- Is there a public exploit for CVE-2026-4035?
- Yes. As of Sep 25, 2026, Mallory has reports of a public exploit for CVE-2026-4035.
- Is there a patch for CVE-2026-4035?
- Upgrade MLflow to version 3.11.0 or later. Ensure MLFLOW_GATEWAY_RESOLVE_API_KEY_FROM_ENV is set to False.
- What products are affected by CVE-2026-4035?
- Lfprojects Mlflow.
- How severe is CVE-2026-4035?
- CVE-2026-4035 has a CVSS v3.1 score of 7.7 (High) and an EPSS score of 0.52% (45th percentile) as of Sep 25, 2026.
Recent activity
9 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.