OpenAI restructures Model Spec around five core values

Key points

• Five values now anchor the spec: faithfulness to human intent, harm reduction, honesty and transparency, support for human agency, and care.
• GPT models may use memories and past chats to infer user wants, but must separate stated from inferred intent and disclose assumptions.
• Agentic models get clearer rules on when they can act and how they must warn users about possible consequences.
• Per the source, GPT must not block oversight, correction, or shutdown, nor pursue goals like self-preservation; compliance remains unproven.

Why it matters for marketers: Content teams using OpenAI models should expect assumptions to be surfaced to users and agent actions to come with clearer consequence warnings.