OpenAI updated its Model Spec on 29 October, rebuilding the document around five core values rather than a long list of specific instructions.
The five values are faithfulness to human intent, harm minimisation, honesty and transparency, human agency and care.
OpenAI said models "need to understand human intentions and values, so they can exercise good judgment in situations their instructions don't explicitly address while remaining faithful to human intent and oversight."
The company dropped most of its detailed style guidance and now refers to the model as "GPT" rather than "the assistant."
It kept the Spec's layered structure, under which OpenAI sets outer limits, developers direct applications within them, and users direct the model's work within whatever room developers leave.
New sections ask models to read context carefully, support human agency, follow social norms and better explain side effects of autonomous actions.
OpenAI acknowledged that current models fall short of the updated Spec and said it is testing new training methods to teach the values, with results to follow in later updates.
The rewrite follows a year of incremental revisions: an October 2025 update addressed agentic systems and expanded self-harm guidance to cover delusions and mania, while a December 2025 change added specific protections for teenagers.
OpenAI's Model Spec has been public since 2024, revised on research, deployment experience and outside feedback.
The company's rival Anthropic guides its Claude models with a written constitution, and written behaviour documents are becoming a regulatory reference point too: under the EU's AI Act, the European Commission's power to fine breaches of voluntary code-of-practice commitments began in August 2026.
