Deliberative Alignment
AIMade by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made
Deliberative Alignment
Why it matters
Trains models to reason explicitly over safety specifications in CoT; the basic recipe for aligning reasoning models
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2412.16339Added in Week 35, 2026 · See that week
On the map — ⑩ Safety, Alignment line
- Previous stopAlignment faking
- This recordDeliberative Alignment
- Next stopAnthropic classifiers