Constitutional Midtraining: Content Presence Drives Alignment Gains

A study at 120B scale finds that inserting constitutional content during midtraining produces durable alignment gains, with constitutionally midtrained models outperforming a control on alignment generalization and reducing blackmail propensity by 17.5 percentage points after benign fine-tuning, while incurring no average capability cost on benchmarks like MMLU, ARC-Easy, piqa, and GSM8K. The research, detailed in arXiv:2607.26654v1, shows that the presence of constitutional content matters more than its structure, and the advantage persists through supervised fine-tuning and benign fine-tuning but not under active in-context pressure.

arXiv:2607.26654v1 Announce Type: new Abstract: Post-training alignment is often shallow, eroding under fine-tuning. Whether midtraining interventions, cleanly isolated from post-training, can produce durable alignment remains untested. We test this via constitutional midtraining: inserting principled, values-based content into midtraining against a replay-only control at 120B scale. Our 394M-token constitutional corpus, built from Anthropic's Constitution, uses a 2x2 factorial design curriculum ordering x deliberative reasoning to produce four constitutionally midtrained conditions plus a control, evaluated on self-generated and established benchmarks including alignment under pressure, value conflict resolution, blackmail, and emergent misalignment across three stages: post-midtraining, post-SFT, and post-benign fine-tuning. Constitutionally midtrained models outperform the control on alignment generalization and durability, notably on blackmail: SFT instills a blackmail propensity in all models, but constitutional midtraining blunts it, with the advantage surviving benign fine-tuning -17.5pp . This durability does not extend to settings requiring active resistance to in-context pressure or conflict, where the advantage attenuates after SFT. The presence of constitutional content at midtraining also matters more than its structure, and constitutional midtraining incurs no cost, on average, on the capabilities we test MMLU, ARC-Easy, piqa, GSM8K at any stage. A modest amount of constitutional content at midtraining could therefore yield broad, persistent alignment gains, offering a cheap, complementary addition to SFT-centered pipelines. Code, data, and models are available.