← Back to all articles
arXiv cs.CLAugust 19, 2026

Constitutional Midtraining: Content Presence Drives Alignment Gains

Excerpt

arXiv:2607.26654v3 Announce Type: replace Abstract: Post-training alignment is often shallow, eroding under fine-tuning. It remains untested as to whether constitutional midtraining interventions can produce durable alignment when cleanly isolated from post-training. We build a 394M-token constitutional corpus from Anthropic's Constitution and apply constitutional midtraining at 120B scale, where principled, values-based content is inserted into midtraining. A 2x2 design (curriculum ordering x d