OpenAI is reorganizing its Model Behavior team, a mini but influential group of researchers that style really nan company’s AI models interact pinch people, TechCrunch has learned.
In an August memo to unit seen by TechCrunch, OpenAI’s Chief Research Officer Mark Chen said nan Model Behavior squad — which consists of astir 14 researchers — would beryllium joining nan Post Training team, a larger investigation group responsible for improving nan company’s AI models aft their first pre-training.
As portion of nan changes, nan Model Behavior squad will now study to OpenAI’s Post Training lead Max Schwarzer. An OpenAI spokesperson confirmed these changes to TechCrunch.
The Model Behavior team’s founding leader, Joanne Jang, is besides moving connected to commencement a caller task astatine nan company. In an question and reply pinch TechCrunch, Jang says she’s building retired a caller investigation squad called OAI Labs, which will beryllium responsible for “inventing and prototyping caller interfaces for really group collaborate pinch AI.”
The Model Behavior squad has go 1 of OpenAI’s cardinal investigation groups, responsible for shaping nan characteristic of nan company’s AI models and for reducing sycophancy — which occurs erstwhile AI models simply work together pinch and reenforce personification beliefs, moreover unhealthy ones, alternatively than offering balanced responses. The squad has besides worked connected navigating governmental bias successful exemplary responses, and helped OpenAI specify its stance connected AI consciousness.
In nan memo to staff, Chen said that now is nan clip to bring nan activity of OpenAI’s Model Behavior squad person to halfway exemplary development. By doing so, nan institution is signaling that nan “personality” of its AI is now considered a captious facet successful really nan exertion evolves.
In caller months, OpenAI has faced accrued scrutiny complete nan behaviour of its AI models. Users powerfully objected to personality changes made to GPT-5, which nan institution said exhibited little rates of sycophancy but seemed colder to immoderate users. This led OpenAI to reconstruct entree to immoderate of its bequest models, specified arsenic GPT-4o, and to merchandise an update to make nan newer GPT-5 responses consciousness “warmer and friendlier” without expanding sycophancy.
Techcrunch event
San Francisco | October 27-29, 2025
OpenAI, and each AI exemplary developers, person to locomotion a good statement to make their AI chatbots friends to talk to, but not sycophantic. In August, nan parents of a 16-year-old boy sued OpenAI complete ChatGPT’s alleged domiciled successful their son’s suicide. The boy, Adam Raine, confided immoderate of his suicidal thoughts and plans to ChatGPT (specifically a type powered by GPT-4o), according to tribunal documents, successful nan months starring up to his death. The suit alleges that GPT-4o grounded to push backmost connected his suicidal ideations. Notably, nan Model Behavior squad did not beryllium erstwhile GPT-4o was developed.
Jang announced successful a post connected X last week that she’s leaving nan squad to “begin thing caller astatine OpenAI.” The erstwhile caput of Model Behavior has been pinch OpenAI for astir 4 years, and antecedently worked connected projects specified arsenic Dall-E 2, OpenAI’s early image procreation tool.
Jang told TechCrunch she will service arsenic nan General Manager of OAI Labs, which will study to Chen for now. However, it’s early days, and it’s not clear yet what those caller interfaces will be, she said.
“I’m really excited to research patterns that move america beyond nan chat paradigm, which is presently associated much pinch companionship, aliases moreover agents, wherever there’s an accent connected autonomy,” said Jang. “I’ve been reasoning of [AI systems] arsenic instruments for thinking, making, playing, doing, learning, and connecting.”
When asked whether OAI Labs will collaborate connected these caller interfaces pinch erstwhile Apple creation main Jony Ive — who’s now working pinch OpenAI connected a family of AI hardware devices — Jang said she’s unfastened to tons of ideas. However, Jang said she’ll apt commencement pinch investigation areas she’s much acquainted with.