# Labs Govern Model Behavior Through Published Values, Not Rulebooks

Frontier labs will keep replacing detailed behavior rules with short published value hierarchies that the model is expected to weigh itself. Their public behavior specs will converge on a constitution-style format, and regulators, auditors and enterprise buyers will increasingly have to assess vendors on how the model's judgment is specified rather than on lists of rules it follows.

- Conviction: 35 / 100 (forming)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-11-01T00:00:00.000Z
- Last updated: 2026-11-01T06:34:19.008Z
- Canonical: https://polylog.news/ai/trends/labs-codify-model-values-over-rules
- Publisher: Polylog
- Affected regions: United States

## Recent evidence

- [confirms] OpenAI rebuilds its Model Spec around five values and retires most of its detailed style rules (2026-11-01): OpenAI's October 29 Model Spec revision retires most detailed style rules and tells the model to weigh five values (faithfulness to intent, harm minimization, honesty, human agency, care) where no explicit rule applies. It also renames the model from "the assistant" to "GPT".