Signal2026-08-13
arXiv

Convergent Detour Hijacking: Task-Preserving Resource Amplification in Skill-Based LLM Agents

Part of

Adaptive Interfaces And AI Benchmarking

Open primary sourceRead on Substack