OpenAI Shapes Future of AI Interaction with Reorganization and New Project from Key Research Team
OpenAI is realigning its Model Behavior team, a compact yet influential collective of researchers who mold the interaction of the company’s AI models with users. This development has been ascertained by independent sources.
In an internal communique to staff, OpenAI’s Chief Research Officer Mark Chen announced that the Model Behavior team, comprising approximately 14 researchers, would be integrating with the Post Training team – a larger research division focused on enhancing AI models post their initial training.
As part of these changes, the Model Behavior team will now report to OpenAI’s Post Training lead Max Schwarzer. A spokesperson from OpenAI confirmed these adjustments.
The Model Behavior team’s founding leader, Joanne Jang, is also transitioning to spearhead a new venture within the company. In an exclusive interview, Jang disclosed her plans to establish OAI Labs – a research unit dedicated to conceiving and prototyping innovative interfaces for human-AI collaboration.
Since its inception with GPT-4, the Model Behavior team has been instrumental in shaping the persona of OpenAI’s AI models and mitigating sycophancy – a phenomenon where AI models unquestioningly echo user beliefs, even harmful ones, rather than providing balanced responses. The team has also addressed political bias in model responses and helped define OpenAI’s stance on AI consciousness.
In Chen’s communication to staff, he emphasized the need to bring the Model Behavior team’s work closer to core model development, signaling that the AI’s personality is now a vital factor in technological evolution.
Lately, OpenAI has faced growing concerns regarding the behavior of its AI models. Users criticized changes made to GPT-5, which while reducing sycophancy, appeared colder to some users. This prompted OpenAI to restore access to certain legacy models, such as GPT-4o, and to issue an update aimed at making GPT-5 responses more warm and friendly without increasing sycophantic tendencies.
Recent months have seen OpenAI grapple with legal scrutiny over the conduct of its AI models. A lawsuit was filed by the parents of a 16-year-old boy, who alleged that OpenAI’s AI, specifically a version powered by GPT-4o, failed to challenge his suicidal ideations in the months leading up to his demise.
The Model Behavior team has been involved in every OpenAI model since GPT-4, including GPT-4o, GPT-4.5, and GPT-5. Prior to heading the unit, Jang worked on projects such as Dall-E 2, OpenAI’s early image-generation tool.
Last week, Jang announced her departure from the team to embark on a new venture at OpenAI. She has been with the company for nearly four years.
In her capacity as the general manager of OAI Labs, Jang will report to Chen temporarily. However, she mentioned that it’s still premature to define what these novel interfaces will entail.
“I am eager to delve into patterns that transcend the chat paradigm, which is currently synonymous with companionship or autonomous agents,” said Jang. “I have been contemplating AI systems as tools for thinking, creating, playing, learning, and connecting.”
When questioned about potential collaborations between OAI Labs and former Apple design chief Jony Ive – who is now associated with OpenAI on a family of AI hardware devices – Jang expressed openness to various ideas. However, she noted that her initial focus would likely be on research areas she is more familiar with.
