Exploring two questions - Do honeypots catch what LLM-based monitors miss? Does knowing honeypots exist allow an agent to evade detection and if so, at what cost to its attack capability?