m00dy.sh
Menu
Search
Home/Projects/The Unchecked Atlas
Project 0436 Cases · v0.2

The Unchecked Atlas

Interactive technical atlas of recurring parser and deserialization failure patterns across AI/ML infrastructure.

Visit the project siteExplore the work
unchecked.sh
The work

Research focus.

Connects public vulnerability cases, untrusted wire fields, memory-safety primitives, and source-anchored defensive guard comparisons across 27+ AI and ML codebases without exposing private research records.

From the notebook

Read into the work.

4 essays

  1. Structure-Aware Fuzzing for ML Model Parsers

    6 min read

  2. Found by AI, Fixed by AI

    7 min read

  3. When an HNSW Index Lies About Its Capacity

    6 min read

  4. The Illusion of Safe Tensors: Why Valid Model Files Still Need Numerical Checks

    17 min read
← All projects