Openai is reorganizing the Model Action Team, a small but influential group of researchers that shape how their AI models interact with people, TechCrunch learned.
In an August memo to staff seen by TechCrunch, Openai’s Chief Research Officer Mark Chen said the model behavior team, which consists of around 14 researchers, will be participating in the post-training team, a large research group responsible for improving AI models after initial training.
As part of the change, the Model Action Team will report to Openai’s post-training lead Max Schwarzer. An Openai spokesperson confirmed these changes to TechCrunch.
Joanne Jang, founding leader of the Model Behavior team, is also on the way to launch a new project with the company. In an interview with TechCrunch, Jang says he is building a new research team called OAI Labs. It is responsible for “inventing and prototyping new interfaces for the way people work with AI.”
The Model Behavioral Team has become one of Openai’s leading research groups responsible for shaping the character of the company’s AI model and reducing sicophany, which occurs when AI models agree and even strengthen unhealthy, rather than providing balanced answers. The team also worked to navigate political bias in model responses and helped Openai define an attitude about AI awareness.
In a note to staff, Chen said it is time to bring the work of the open model action team closer to core model development. In doing so, the company shows that AI’s “personality” is considered a key factor in the evolution of technology.
Over the past few months, Openai has faced an increasing scrutiny of AI models’ behavior. Users strongly opposed the personality changes made to GPT-5. The company said it had a low percentage of psychofancy, but it appears to be cold for some users. This has restored access to some legacy models such as the GPT-4O and released updates to ensure that the new GPT-5 response feels “warm and friendly” without increasing psychofancy.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
Openai and all AI model developers need to walk the fine lines to make AI chatbots friendly and talk rather than sikopantic. In August, the parents of a 16-year-old boy sued Openai over the suspected role of ChatGpt in his son’s suicide. Boy Adam Lane opened up some of his thoughts on suicide, and in the months leading up to his death, he planned a chat GPT (particularly the version equipped with the GPT-4o) in the months leading up to his death. The lawsuit alleges that GPT-4o could not push back his suicidal ideation.
The Model Action Team has worked on all OpenAI models since GPT-4, including GPT-4O, GPT-4.5 and GPT-5. Before starting the unit, Jang previously worked on projects such as Dall-E 2, Openai’s early image generation tool.
Last week, Chang announced in an X post that she would leave the team and “start something new in Openai.” The former head of model action was in Openai for almost four years.
Chan told TechCrunch that he will be general manager of OAI Labs and will report to Chen for now. But it was an early age and it’s not yet clear what will happen with those novel interfaces, she said.
“I’m really excited to explore the patterns that move us beyond the chat paradigms that are currently associated with dating and agents. “I’ve been thinking about it. [AI systems] As an instrument for thinking, creating, playing, doing, learning, and thinking about connections. ”
When asked if OAI Labs would work with former Apple design chief Jony Ive on these new interfaces, he is currently working with Openai on a family of AI hardware devices, but Jang said he is embracing many ideas. But she said she would probably start with a field of research she knows better.
This story has been updated to include a link to Chan’s posts announcing her new positions that were released after the story was released. It also clarifies the models that Openai’s model action team has worked on.
Source link
