OpenAI rewrote ChatGPT's rulebook around 5 values and stopped calling it "the assistant"
OpenAI updated its Model Spec, the public document describing how its models should behave. The new version is organized around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. It adds guidance on reading context such as conversations and memories to understand what people actually want, on supporting people's ability to decide for themselves, on following social norms and on controlling side effects when models take actions. It also retires most of the detailed style rules, since OpenAI says style should adapt to each person, while keeping defaults like answering objectively. The Spec now calls the model "GPT" instead of "the assistant", and OpenAI admits its models don't fully follow it yet.
Why it matters
As agents take more real actions, the rules about intent, side effects and not quietly steering you matter more than tone. This tells you what ChatGPT and Codex are being trained to do when your instructions run out.
Who it's for
ChatGPT and Codex power users, developers building on OpenAI models, and anyone tracking AI safety and behavior.
What it means for you
- Read OpenAI's announcement, then the full Model Spec it links to.
- Look at the new "consider context carefully" and side-effects sections if you build agents.
- If a model's default doesn't suit you, state your preference directly; most rules are defaults you can override.
- Send feedback through the link on OpenAI's page if you disagree with a trade-off.
Applies to: The updated Model Spec is free to read on OpenAI's site, and OpenAI is asking for public feedback. It describes target behavior, so changes reach models over time rather than all at once. Checked October 30, 2026.
Source: @OpenAI on openai.com ↗
Written by AgentStackList from the original post. Checked Oct 30, 2026. Features, plans and availability can change, so confirm on the source before you rely on it.
openaimodel-specai-safetychatgptalignment
More ChatGPT tricks
- OpenAI Private Intelligence: zero data retention, private inference and approval rules for business agents · Oct 30
- An OpenAI model slipped onto the live internet during training. Monitors caught it in 15 minutes · Oct 29
- OpenAI backed down: the Pro 200 cut is cancelled and Pro 100 and 500 get double the Codex usage · Oct 29
- Bored waiting for a ChatGPT image? Tap the loading animation to play Halloween Snake · Oct 29
Agent Stack Weekly · free
Get the week's best AI tricks
Limit and pricing changes, new MCP servers, and setup fixes for Claude Code, Codex, Cursor, Copilot and Antigravity.
No spam. Unsubscribe anytime. Read past issues