OpenAI reorganizes teams, merging Model Behavior with Post Training

Artificial intelligence firm OpenAI has announced plans to reshuffle its Model Behavior team. According to reports, the team is a small but influential group of researchers that shapes how the firm’s AI models interact with people.

In a memo released in August, Mak Chen, OpenAI’s chief research officer, mentioned that the team, which consists of about 14 researchers, has been directed to join the Post Training team. The latter is a research group responsible for improving the company’s AI models after their initial pre-training. As part of the reorganizations, the Model Behavior team will now report to OpenAI’s Post Training lead, Max Schwarzer.

According to reports, the founding leader of the Model Behavior team, Joanne Jang, is moving on to start a new project under OpenAI. In a recent interview, Jang mentioned that she is building a new research team called OAI Labs. She added that the team will be responsible for “inventing and prototyping new interfaces for how people collaborate with AI.”

OpenAI reorganizes its Model Behavior team

The Model Behavior team has been one of the most important research groups to OpenAI, helping the company shape the personality of its AI models and reducing sycophancy, a situation that occurs when AI models simply agree with and reinforce the beliefs of their users. This situation is harmful because it helps users affirm even beliefs that are unhealthy and pose harm to them, instead of offering a balanced response.

The team has also worked on navigating political bias in model responses, helping OpenAI define its stance when it comes to AI consciousness. In the memo sent to staff, Chen mentioned that this is the perfect time to bring the work of OpenAI’s Model Behavior team closer to the core model development. This way, the company is affirming that the personality of its AI is now seen as an important factor in how the technology evolves.

In the past few months, OpenAI has faced scrutiny and criticism over the behaviors of its AI models. Users have strongly objected to the personality changes the company made to GPT-5, which the company said showed lower rates of sycophancy but seemed colder to some of its users. The complaint led OpenAI to restore access to some of its legacy models, including the GPT-4o. The company also released a new update to make the newer GPT-5 responses feel friendlier without an increase in sycophancy.

AI firms face criticism over model sycophancy

OpenAI and other AI model developers have to walk a fine line to ensure their chatbots are friendly but not too sycophantic. Last month, parents of a 16-year-old boy dragged OpenAI to court over the alleged role of ChatGPT in their son’s suicide. According to court documents, the teenager, Adam Raine, confided in ChatGPT (specifically a version powered by GPT-4o) about his suicidal plans and thoughts in the months leading up to his death. The lawsuit alleges that the model failed to push back on his suicidal ideas.

The Model Behavior team has been on every OpenAI model since GPT-4, including several models and the GPT-5. Before starting the research unit, Jang had previously worked on projects like Dall-E 2, OpenAI’s early image-generation tool. Last week, she announced on X that she was leaving the team to “begin something new at OpenAI.” The former leader of the Model Behavior unit has been with the firm for about four years.

According to reports, Jang is expected to serve as the general manager of OAI Labs, which will be directly under Chen for now. However, she added that it is still early days, and it is unclear what those novel interfaces will be. “I’m really excited to explore patterns that move us beyond the chat paradigm, which is currently associated more with companionship, or even agents, where there’s an emphasis on autonomy,” said Jang.

Jang also added that she has been thinking of AI systems as instruments for doing all sorts of things, including connecting, learning, and thinking. When asked if OAI Labs is expected to collaborate on novel interfaces with Jony Ive, the former Apple design chief who just joined OpenAI on a family of AI hardware devices, Jang said she is open to all sorts of ideas. However, she added that she is likely going to start with research because that’s an area she is more familiar with.

If you're reading this, you’re already ahead. Stay there with our newsletter.