AIsbom Research Blog
Deep-dives, guides, and tutorials on securing the AI model supply chain.
The AIsbom Platform is Generally Available
The hosted dashboard at app.aisbom.io is now open to everyone: a continuous AI Bill of Materials fed by your own CI. Connect a repo, add one input to the GitHub Action, and every PR scan becomes part of a living inventory you can diff, share, and hand to compliance.
Safetensors vs. Pickle: A Security Comparison
Why Python's Pickle format leads to Remote Code Execution by design, how SafeTensors closes the attack surface while loading faster, whether PyTorch's weights_only default is a real fix, and which model supply-chain risks remain even after you migrate.
What is an AI BOM? The AI Bill of Materials, Explained
An AI BOM (AI Bill of Materials) is a structured inventory of the model weights, training data, licenses, and frameworks behind an ML system. Why traditional SBOMs miss it, what to track, how CycloneDX and SPDX represent it, and how to generate one in seconds.
How to Detect Malware in a PyTorch Pickle File: A Hands-On Guide
A deep-dive tutorial explaining how PyTorch model pickles execute code, how exploit payloads work, and how to statically scan them for malware using Python's standard library.