Who May Release an Agent's Promise? Standing over Commitments in Multi-Party Chat
When an assistant serving a team makes a promise to Alice, who may cancel it?
ProofPaper ↗
Key points
- We call this standing: only the creditor (the person owed) or someone the creditor or operator policy authorizes may release a commitment, regardless of rank.
- In a controlled chat environment we vary who cancels and separate two errors: unauthorized release (obeying someone without standing) and override (executing a released promise).
- Oracles dissociate the errors: stating the rule or the creditor curbs unauthorized release but not override, which the latter even raises in three of four models; among them, only knowing whether the promise is still owed cuts override.
- A self-kept commitment ledger stops override in four of five models; with a one-paragraph release policy it matches or beats the best oracle store, raising gpt-6-luna's strict accuracy from 49% (policy alone) to 83%.
Sources (1)
- [1]Who May Release an Agent's Promise? Standing over Commitments in Multi-Party ChatarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 04:01 AM
When an assistant serving a team makes a promise to Alice, who may cancel it?
We call this standing: only the creditor (the person owed) or someone the creditor or operator policy authorizes may release a commitment, regardless of rank.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 15, 2026Local and Distributed Large LLM Inference with MLX and RDMA on Mac Studio
- Oct 15, 2026Terminal-RSI: Recursive Self-Improving Terminal Agents via Recursive Self-Rewrite and Policy-Refine
- Oct 14, 2026ReaDy: Recurrent States as Dynamic Context for Speculative Drafting in Hybrid Linear Attention Large Language Models
- Oct 13, 2026My local Qwen3.8FN agent is significantly cheaper than cloud
- Oct 13, 2026Swift 1.5 vs ThinkingCap 3.8 vs Qwen3.8 27B: Accuracy and Token Efficiency
- Oct 13, 2026PSA for 3090 FE owners: Running LLMs overheats your VRAM