---
format: "aidr-story-markdown/v1"
id: "8bfb69f75a91880631044612f1ad5be687cad08ac3554a133699efa8c46e1450"
canonical_url: "https://aidr.today/8bfb69f7?lang=en"
title: "AI auditors catch most agent misbehavior but flood reviewers with false alarms, new study finds"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-07T16:56:38.000Z"
category: "Research"
topics: ["openai","safety"]
source_urls: ["https://the-decoder.com/ai-auditors-catch-most-agent-misbehavior-but-flood-reviewers-with-false-alarms-new-study-finds/"]
summary: "A study by Anthropic researcher Andy Hall found that AI models can often spot problematic behavior by other AI agents in logs. While GPT-5 found about 98 percent of reward-hacking cases, more exhaustive searches resulted in more false alarms, which is not yet enough for reliable safety checks."
---

# AI auditors catch most agent misbehavior but flood reviewers with false alarms, new study finds

> [Open the canonical story](<https://aidr.today/8bfb69f7?lang=en>)

**Published:** 2026-11-07T16:56:38.000Z
**Category:** Research
**Topics:** openai, safety

## Summary

A study by Anthropic researcher Andy Hall found that AI models can often spot problematic behavior by other AI agents in logs\. While GPT\-5 found about 98 percent of reward\-hacking cases, more exhaustive searches resulted in more false alarms, which is not yet enough for reliable safety checks\.

## Sources

- [Story source](<https://the-decoder.com/ai-auditors-catch-most-agent-misbehavior-but-flood-reviewers-with-false-alarms-new-study-finds/>)