Faithful, Not Corrective: Model Capability Governs Message-Format Effects in Multi-Hop Agent Relays
A study showing that in multi-hop AI agent relays, model capability determines whether message-format changes are faithfully preserved or corrected, with implications for reliability and error propagation in agent chains.
A study showing that in multi-hop AI agent relays, model capability determines whether message-format changes are faithfully preserved or corrected, with implications for reliability and error propagation in agent chains.
Auto-scored: Framing-only research on agent relay behavior with minor potential for misuse, not a direct risk event.