Harshil2301/zer0n-bench

Zer0n-Bench addresses a high-value niche in AI-driven security: the poor quality of existing vulnerability datasets (like Big-Vul or NVD-derived sets). However, with 0 stars, 0 forks, and being only 16 days old, it currently exists as a pre-release or personal research artifact rather than a viable project. Its defensibility is almost non-existent as it lacks community adoption and data gravity. In the competitive landscape of security benchmarks (SARD, Juliet, DiverseVul), Zer0n-Bench's unique selling point is 'provenance-awareness'—tracking why a label exists—to study label noise. While frontier labs are unlikely to build this specific academic tool, the rapid advancement of LLMs for automated labeling and code analysis could render static, manually curated benchmarks like this obsolete within 1-2 years. The primary risk is not platform domination by Google/AWS, but rather displacement by more comprehensive, LLM-generated synthetic datasets that solve the label noise problem through scale and reasoning rather than provenance tracking.