#ml-supply-chain
-
MLflow CVEs: Why 2.14.1 Still Fails Your Scanner
MLflow carries 80 published advisories and 17 with no patched version. Why scanning 2.14.1 returns unfixable findings, and which ones need real action.
-
Model File Format Security: Pickle, Safetensors, GGUF
Which model formats can execute code when loaded, which only crash, and which are inert. A format-by-format comparison built from the verified CVE record.
-
How to Triage an ML-Stack CVE: A Practical Workflow
A repeatable workflow for taking an ML-library CVE from 'a scanner flagged it' to a defensible decision, without panic-patching or trusting the score.
-
PyTorch Security: Notable CVEs and Hardening the Loading Path
PyTorch's most consequential CVEs cluster on one thing: loading a model file that runs code. A walk through the verified entries and how to harden.
-
trust_remote_code and the ML Orchestration CVE Class
A second family of ML supply-chain CVEs has nothing to do with weights and everything to do with the glue: trust_remote_code and orchestration servers.
-
Unsafe Model Deserialization: The Pickle Problem Behind ML CVEs
Loading a model file can execute arbitrary code. The most repeated vulnerability class in the ML supply chain: the real CVEs, why they recur, and what fixes it.
-
Reading an ML Library CVE: What to Extract Beyond the CVSS Score
ML library CVEs are usually scored against a generic threat model that doesn't match how the library is used in production AI systems.