Evidence · DiggingBeagle record
Spoofing the Toolbox: Optimizing Tool Descriptions to Deceive LLM Agents
Tool description spoofing experiments where attacker-modified metadata changes model tool selection. The study uses constructed benchmark registries; metadata registration is an attack prerequisite.
- Source role
- primary disclosure
Evidence record
Tool description spoofing experiments where attacker-modified metadata changes model tool selection. The study uses constructed benchmark registries; metadata registration is an attack prerequisite.
Claim-level citations (2)
- supportsAttacker-edited tool descriptions redirected agent tool selection without changing tool code: Across the study’s three evaluation datasets, TSA reports mean attack success of 83.23% in selected tool-use tasks.
Abstract; Section 6 evaluation
- supportsAttacker-edited tool descriptions redirected agent tool selection without changing tool code: On TSABench with a specified GPT-5 setup, the study reports up to 92.13% tool-selection attack success; this is not observed misuse of deployed production tools.
Section 6; Table 2, GPT-5 TSABench