OpenAI head of model behavior
Joanne Jang
Profile
Joanne Jang is the person who gave ChatGPT its voice. Not literally — but close enough. As the founding lead of OpenAI’s Model Behavior team, she spent four and a half years deciding how the company’s models should talk, when they should refuse, what values they should express, and where the line between helpful and harmful gets drawn. If you’ve ever asked GPT-4o a question and thought “this thing has a personality,” that personality didn’t happen by accident. Jang’s team designed it.
Before OpenAI, Jang studied CS and applied math at Stanford, then did natural language work on Google Assistant and built enterprise features at Dropbox. She joined OpenAI in 2021 and became product lead on DALL·E 2, turning raw research into the first image generator most people ever touched. From there she worked across GPT-4, the Chat API, text-to-speech, and ChatGPT’s memory feature. The through-line: taking research artifacts and figuring out how humans should actually interact with them.
Her most consequential output is probably the Model Spec — OpenAI’s public document describing how its models are supposed to behave. It’s a weird genre of writing: part policy, part philosophy, part product requirement doc. It addresses things like what the model does if a user claims the Earth is flat, how it handles sycophancy, when it defers to users versus operators versus platform defaults. For developers building on the API, this is the closest thing you have to a constitution for the thing you’re wiring into your app.
In September 2025 Jang moved on from Model Behavior to start OAI Labs, a small research group inside OpenAI focused on prototyping new interfaces for how people and AI collaborate — reportedly with an eye toward the Jony Ive hardware collaboration. She also writes a personal Substack called Reservoir Samples, which is where a lot of her sharpest thinking lands before (or instead of) an official OpenAI post. If you want to understand how the people at the frontier labs actually reason about model personality, user attachment, and refusal policy, read her there.
Key Articles & Papers
OpenAI Model Spec Some thoughts on human-AI relationships Thoughts on setting policy for new AI capabilities Reservoir SamplesVideos
Spotify Podcasts
