arXivNeeraj Karamchandani, Piyush Nagasubramaniam, Xinhong Xie, Sencun Zhu, Dinghao WuFri, Oct 2, 2026, 9:55 AM PDT
score 15.1
AI agent security scores change when tool names change
Original: Threat-Preserving Representation Sensitivity in Agent-Security Benchmarks
Source: arxiv.org ↗
Writing ELI5 summary…