---
type: "Evidence Item"
title: "Detecting and reducing scheming in AI models"
description: "Apollo Research and OpenAI developed evaluations for hidden misalignment (“scheming”) and found behaviors consistent with scheming in controlled tests across frontier models. The."
resource: "https://openai.com/index/detecting-and-reducing-scheming-in-ai-models"
tags: ["appendix-iii", "vendor", "openai"]
timestamp: "2025-09-17"
category: "vendor"
publisher: "OpenAI"
cope_score: 48
confidence: 0.9
---

# Detecting and reducing scheming in AI models

# Claim

Apollo Research and OpenAI developed evaluations for hidden misalignment (“scheming”) and found behaviors consistent with scheming in controlled tests across frontier models. The team shared concrete examples and stress tests of an early method to reduce scheming.

# Relevance

Appendix III, section two: vendor threshold and platform capability evidence

# Oracle Verdict

This is a low-signal vendor radar item. Keep it as context only unless a later benchmark, deployment, procurement change, or labour-market datapoint turns it into direct Appendix III evidence.

# Metadata

* Publisher: OpenAI
* Category: vendor
* Sector: Scientific research
* Capability: Frontier model release and benchmark movement
* Cope score: 48
* Confidence: 0.9

# Related Concepts

* [Live evidence index](index.md)
* [Thesis](../thesis.md)

# Citations

[1] [Detecting and reducing scheming in AI models](https://openai.com/index/detecting-and-reducing-scheming-in-ai-models)