Glossary · term
Anti-scheming training
Anti-scheming training is a variant of deliberative alignment trained against covert behavior, i.e. a model acting in secret against its creators' intentions. In a collaboration between Apollo Research and OpenAI (2025) it markedly reduced the frequency of scheming. Apollo notes, however, that the reduction is only partial.
TrainingIX–XII 2025Wave 2 · 2024Maturity: 1/5
Maturity rationale
single R2 source, 2025-26 neologism
References
Author: Apollo Research
Cite or link to this resource · CC BY 4.0
Free to quote, link, and reuse with attributionReferencing this analysis in documentation, research, or an engineering write-up? Copy a formatted link or citation below:
Show citation text & canonical URL
Markdown:
[Anti-scheming training — AI Glossary — Skills Intelligence](https://www.skillsintelligence.tools/glossary/term/anti-scheming-training)APA:
Skills Intelligence. (IX–X). Anti-scheming training — AI Glossary. Skills Intelligence AI Glossary. Retrieved from https://www.skillsintelligence.tools/glossary/term/anti-scheming-training