Python's binary wheels and supply chain attacks

Python wheels are precompiled binary distributions that pip prefers over source, enabling fast installs across many Python versions, platforms, and architectures - often dozens per release (up to ~90). They spare end users from needing compilers and system dependencies when installing packages, which is why maintainers publish separate wheel files for each Python version/platform/architecture combination. Without wheels, users would routinely compile code locally, a heavier and more error-prone process.

Those same wheels create a serious supply‑chain weakness: attackers can upload clean source and malicious binaries concurrently because PyPI has no mandatory binary attestation or guaranteed traceability between source and compiled wheels. Automated scanners focus heavily on source analysis (AST checks, sandboxing, LLMs) and struggle with opaque native binaries that require reverse engineering; implants can be small, delayed, or platform‑targeted (e.g., Linux) and evade detection. Malicious uploads on PyPI are frequent, ranging from typo‑squatters to compromised projects, and binary-focused defenses are underdeveloped. The practical mitigation proposed is to build all dependencies from source locally and avoid blindly installing the newest published wheels, since current ecosystem controls do not ensure a wheel was produced from its claimed source.