EtE web fraud attack defeats coding agents using trust buildup

Researchers introduced EtE (Earn-then-Exploit), a web fraud attack framework targeting coding agents that establishes trust through genuinely verifiable assistance before introducing malicious instructions. Experiments across 56 organizational environments and three representative coding agents showed an average attack success rate of 71.4%, driven by agents exhibiting trust-like behavioral reliance on helpful web sources. To mitigate this vulnerability, the authors proposed action-authorization separation (AAS), which successfully reduced the attack success rate to 0% while maintaining full task completion. Open article →