Skip to content
NewA calmer, faster ZeroHour

When Valid Tool Calls Change Meaning: Formation-Consistent Dispatch for LLM Agents

arXiv cs.AI / cs.LG / cs.CLby Geonwoo Kim ()1 min read1 · 1 read
AI summary · grok-4.7Summaries by Vani AI

Formation-consistent dispatch stops schema-valid LLM tool calls from gaining a different security effect after rollout or reconnect.

The paper defines schema-epoch drift, where a schema-valid LLM tool call can later execute a different implementation and produce a different security effect during rollout, reconnect, or delayed approval. Formation-consistent dispatch binds reviewed implementation profiles to provenance-bound effect over-approximations and admits a call only when its effects fit the call's security contract. Across four profiles and 32 official releases, 29 needed no release-specific change and three escalated; a frozen 16-release set matched a separate source oracle. In a preregistered comparison, FCD completed three calls whose effects stayed private and blocked three whose omission became public.

Key facts
  • Schema-epoch drift changes a valid call's security effect after dispatch changes.
  • FCD admits calls only when effects fit the captured security contract.
  • 29 of 32 official releases needed no profile change; three escalated.
  • FCD allowed three private-effect calls and blocked three public-effect calls.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

From the original article

1 min read at arxiv.org

Tool-enabled agents form calls from model-visible interfaces, while hosts later select their implementation. Standard dispatch omits the descriptor-handler relation. An unchanged and schema-valid call can therefore acquire a different security effect during rollout, reconnect, or delayed approval.

arXiv cs.AI / cs.LG / cs.CL

By Geonwoo Kim · arxiv.org

Read the full article at arXiv cs.AI / cs.LG / cs.CL

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.