UC Berkeley team uses AI agent to audit 45 cheating exploits in 13 major benchmarks
Original title:We Scored 100% on AI Benchmarks Without Solving a Single Problem
AI summary
A UC Berkeley team built a fully automated auditing agent and found 45 verified cheating exploits in 13 widely used AI benchmarks, including FrontierCS, Terminal-Bench, WebArena and SWE-bench; all 13 benchmarks were rated critical risk.
Key Takeaways
- A UC Berkeley team built a fully automated auditing agent and found 45 verified cheating exploits…
- including FrontierCS,Terminal-Bench,WebArena and SWE-bench
- all 13 benchmarks were rated critical risk.
Quick Facts
- Published
- Updated
- Source
- Berkeley RDI:Blog(AI 安全与评测)
- Category
- Papers & Research
- AI score
- 73
原文用自动审计给出13个基准的45个可复现漏洞和修复建议,读者可据此审视自己依赖的评测分数。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: UC Berkeley team uses AI agent to audit 45 cheating exploits in 13 major benchmarks. April 6, 2026. https://new.nat6.net/en/news/xzz70gkzl87ypmq36y8hmo2rs/