All articles
A focused tracker for CVEs in ML and AI infrastructure. PyTorch, TensorFlow, ONNX, vLLM, llama.cpp, transformers, langchain, LlamaIndex, model registries, and the broader AI/ML supply chain — dated, sourced to NVD or vendor advisory.
-
CVE Tracking Tool for ML Stacks: What to Run in 2026
Which CVE tracking tool to run on an ML stack: OSV-Scanner or pip-audit in CI, Dependency-Track for SBOMs, KEV and EPSS to rank fixes.
-
AI SBOM for ML Models: What Belongs in One
The G7 and CISA minimum elements for an AI SBOM, the 13 model-specific fields they define, and how CycloneDX ML-BOM and SPDX 3.0 represent them.
-
ML Pipeline Supply Chain Attack: From Poison to Production
An ML pipeline supply chain attack turns trusted models, datasets, dependencies, or build artifacts into code execution or persistent model backdoors.
-
AI Bill of Materials Tools: AIBOM Formats and Generators
Two AIBOM standards (CycloneDX ML-BOM, SPDX 3.0 AI profile), one free OWASP generator, and how to pick a tool without buying shelfware.
-
AI CVEs in CISA KEV: Langflow, LiteLLM, Ray, and the Gaps
Six AI-stack CVEs sit in CISA KEV: three Langflow, two LiteLLM, one Ray. What confirmed exploitation looks like, and why ShadowRay is not listed.
-
Inference Server CVEs: vLLM, Ollama, llama.cpp, Triton
The serving layer now outproduces model files for critical CVEs. Verified entries in vLLM, Ollama, llama.cpp and Triton, and the three defects behind them.
-
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.
-
ML Model CVE Tracking: How to Monitor Your AI Stack
How to track CVEs across ML frameworks, inference servers, and model artifacts, and what to do when a vendor disputes the advisory anyway.
-
Malicious Model File Detection: Auditing ML Models
Pickle serialization flaws, PickleScan bypass CVEs, and a practical detection stack for teams pulling models from public repositories like Hugging Face.
-
ML Vulnerability Scanner Software: Static vs Dynamic Tools
A buyer's guide to ML vulnerability scanner software, comparing static model-artifact scanners like ModelScan and fickling with dynamic tools like garak.
-
Hugging Face Model Supply Chain Risk: Pickle Backdoors
How Hugging Face model supply chain risk works: pickle backdoors, the Transformers RCE CVE cluster, why the Hub scanner misses them, and what cuts risk.
-
TensorFlow Security Vulnerabilities 2026: CVEs and Fixes
TensorFlow and Keras security in 2026: CVE-2025-49655, CVE-2025-12058, one TensorFlow 2.18.0 DoS flaw, and model supply-chain risks.
-
LangChain Security Vulnerabilities 2026: CVEs and Fixes
Six verified CVEs in LangChain and LangGraph expose secrets, files and chat history; one chain reaches RCE. CVSS scores, attack paths, patched versions.
-
Best AI Supply Chain Security Tools in 2026
A practitioner's guide to AI supply-chain security tools: model artifact scanners, MLOps pipeline hardening, AIBOM generators, and where each one fits.
-
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: CVEs and Model Loading Hardening
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: Pickle CVEs and Defenses
Pickle-based model loading can execute arbitrary code. CVE-2025-32434 shows why safer formats, provenance checks, scanning, and sandboxing matter.
-
ML CVE Database: What's Tracked and What's Missing
How NVD and MITRE track ML framework CVEs, why AI-specific risks fall outside the CVE taxonomy, and the monitoring layers that close the gaps.
-
ML Library CVE Analysis 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.