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.
ChatGPT
ARTICLE
17 September 2025
Detecting and reducing scheming in AI models
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.
More from ChatGPT
All updatesNew usage analytics and updated spend controls for enterprises
Open article →
Improving health intelligence in ChatGPT
Open article →
Using AI to help physicians diagnose rare genetic diseases affecting children
Open article →
A near-autonomous AI chemist improves a challenging reaction in medicinal chemistry
Open article →
Comments
Sign in or join free to leave a comment.
No comments yet. Be the first.