OpenAI reorganizes research team behind ChatGPT’s personality

OpenAI reorganizes its model behavior team, a small but influential group of researchers who shape how the company’s AI models deal with people, has learned WAN.

In a memo in August for the staff that was seen by TechCrunch, Openai’s Chief Research Officer Mark Chen said that the model behavioral team-that consists of around 14 researchers come to the Post Training Team, a larger research group that is responsible for improving the AI ​​models of the company after their first training.

As part of the changes, the model behavior team will now report to OpenAI’s Post -Training Lead Max Schwarzer. A spokesperson for OpenAi confirmed these changes to WAN.

The founder of the model behavior team, Joanne Jang, also continues to start a new project at the company. In an interview with WAN, Jang says that she is building a new research team called OAI Labs, which will be responsible for “inventing and prototating new interfaces for how people work with AI.”

The model behavior team has become one of the most important research groups of OpenAi, responsible for shaping the personality of the AI ​​models of the company and for reducing sycophancy -which occurs when AI models simply agree with and strengthen user beliefs, even unhealthy, rather than offering balanced answers. The team has also worked on navigating political bias in model reactions and has helped to define its position AI -Consciousness.

In the memo of the staff, Chen said that it is now time to bring the work of the model of the OpenAI model closer to the development of the core model. By doing this, the company indicates that the ‘personality’ of his AI is now considered a critical factor in how the technology evolves.

In recent months, OpenAi has had to deal with increased control over the behavior of its AI models. Users had a strong objection to personality changes that were made in GPT-5, the company said they had a lower percentages of Sycofancy, but for some users seemed colder. This led OpenAi to restore access to some of its old models, such as GPT-4O, and to release an update to make the newer GPT-5 reactions feel “warmer and friendlier” without increasing the sycofancy.

OpenAi and all AI model developers must run a fine line to make their AI chatbots friendly to talk to, but not Sycophantic. In August are the parents of a 16-year-old boy Monkeys Openai charged About the alleged role of Chatgpt in the suicide of their son. The boy, Adam Raine, trusted some of his suicide thoughts and plans to chatgpt (in particular a version driven by GPT-4O), according to judicial documents, in the months prior to his death. The lawsuit claims that GPT-4O has not succeeded in reducing his suicidal thoughts.

Since GPT-4, the model behavior team has worked on every OpenAI model, including GPT-4O, GPT-4.5 and GPT-5. Before Jang started, Jang previously worked on projects such as Dall-E 2, OpenAi’s early tool for image generation.

Jang announced in a Post on X Last week she leaves the team to ‘start something new at OpenAI’. The former head of model behavior has been at OpenAI for almost four years.

Jang told WAN that she will serve as general manager of OAI Labs, who will report to Chen for the time being. However, it is early days, and it is not yet clear what those new interfaces will be, she said.

“I am very enthusiastic about exploring patterns that go beyond the chat paradigm, which is currently more associated with company, or even agents, where the emphasis is on autonomy,” Jang said. ‘I thought about it [AI systems] As instruments for thinking, making, playing, doing, learning and connecting. “

When asked whether OAI Labs will work together on these new interfaces with former Apple Design Chief Jony Ive – Who is now at work With OpenAI on a family of AI -hardware devices -Jang said she is open to many ideas. However, she said she will probably start with research areas that she is more familiar with.

This story was updated with a link to Jang’s message in which her new position was announced, which was released after this story was published. We also clarify the models on which OpenAi’s model behavior team has worked.