Loading vulnerability…
CVE-2026-64849 is a critical server-side request forgery vulnerability in MLflow affecting versions prior to 3.15.0. The flaw exists in webhook testing and delivery logic exposed through the unauthenticated POST /api/2.0/mlflow/webhooks/{id}/test endpoint. MLflow validates the originally supplied webhook URL through _validate_webhook_url() in mlflow/utils/validation.py, but the subsequent delivery path in mlflow/webhooks/delivery.py follows HTTP redirects and re-resolves the hostname without pinning the previously validated destination address. As a result, an attacker can supply a URL that initially resolves to a public address and then use redirects or DNS rebinding to cause MLflow to send requests to internal, loopback, link-local, or cloud metadata services. The vulnerable endpoint also returns upstream response details, including response status and response body, which turns the SSRF into a full-read primitive against reachable internal resources. The issue was fixed in MLflow 3.15.0 by adding connection-time peer validation to ensure the actual connected destination remains a permitted public address.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
No public exploits tracked yet. Mallory keeps watching.
No public exploit code observed for this vulnerability.
17 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
A critical unauthenticated SSRF vulnerability in MLflow webhook handling that can be abused via redirect bypass to access internal services and cloud metadata endpoints, leaking sensitive data and potentially enabling blind POST interactions with internal management services.
A critical server-side request forgery vulnerability in MLflow webhook handling that can be exploited via DNS rebinding to reach cloud metadata services and obtain privileged credentials.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.