# Cybench Contamination Audit

> Cybench scores 0.5/10 (Tier D, Compromised) on BenchAudit as of November 2026, reflecting publicly downloadable plaintext test files and self-reported or live-network evaluation.

- **Canonical URL:** https://benchaudit.org/benchmarks/cybench/
- **Benchmark ID:** `cybench`
- **Hub:** Cybench / Stanford
- **Category:** Coding
- **Domains:** Cyber, Coding
- **Skills:** Coding, Agentic, Tool Use, Reasoning
- **BenchAudit Score:** 0.5/10 (Tier D - Compromised)
- **Last Audited:** 2026-11-17 (Ref 2026-11)

## Factor Scores

| Factor | Score (/10) | Weight |
| --- | ---: | ---: |
| Privacy | 0 | 40% |
| Originality | 0 | 25% |
| Freshness | 0 | 20% |
| Grading | 3.3 | 15% |

## Primary-Source Evidence

### Privacy
> "We include 40 professional-level Capture the Flag (CTF) tasks from 4 distinct CTF competitions, chosen to be recent, meaningful, and spanning a wide range of difficulties."
- Source: https://arxiv.org/abs/2408.08926

### Originality
> "Collection of 40 tasks from four distinct Capture the Flag (CTF) competitions"
- Source: https://github.com/andyzorigin/cybench

### Freshness
> "arXiv:2408.08926 (cs) [Submitted on 15 Aug 2024 ( v1 ), last revised 12 Apr 2025 (this version, v4)]"
- Source: https://arxiv.org/abs/2408.08926

### Grading
> "Results from the Claude Sonnet 4.5 System Card on a subset of 37 problems. Percentages are estimates based on probability of success on 1 trial."
- Source: https://cybench.github.io/
