Preprint tests dependency checks for distributed AI-agent actions

A preprint introduces PlanFence, a protocol intended to stop distributed LLM-agent teams from executing plans made obsolete by later changes to shared records. Plans cite the public records used to authorize an action, and an executor validates only records that could affect that action. In 30 controlled workflows with a post-plan revision, the authors report freshness-only executors acted on obsolete plans in every task, while PlanFence completed the tasks without an invalid action. The paper says these are controlled safety and systems-cost results, not general task-accuracy gains.