# Cheap open-weight surrogate predicts bad agent tool calls at 0.825 AUROC where the agent's own stated confidence scores 0.598

A companion paper finds that one message-passing protocol between LLM agents raises accuracy on ContextBench but lowers it on ToolSandbox, from 0.889 to 0.653. A better summary can still hurt the agent that receives it.

- Published: 2026-10-06T06:03:25.241Z
- Canonical: https://polylog.news/ai/2026-10-06/cheap-open-weight-surrogate-predicts-bad-agent-tool-calls-at
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.AI](https://arxiv.org/abs/2610.03894), [arXiv cs.LG](https://arxiv.org/abs/2610.03769)

Agents built on frontier chat APIs produce tool calls, queries and code that can be wrong with no visible sign. The error usually appears only after the action has run. Proxy Confidence, by Yikai Zhao, Saurabh Pandey and Pradeep Kumar Misra…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-10-06/cheap-open-weight-surrogate-predicts-bad-agent-tool-calls-at (subscription information: https://polylog.news/pricing).