arXivWei-Lin Chen, Zhepei Wei, Xinyu Zhu, Yu-Min Tseng, Tzu-Han Lin, Yu MengMon, Nov 2, 2026, 10:27 AM PST
score 17.0
AI coding agents often succeed while causing hidden harm
Original: VeriHarm-Bench: Measuring Harmful Success in LLM Coding Agents Verifiably
Source: arxiv.org ↗
Writing ELI5 summary…