What to know

  • The small Model Behavior team (about 14 researchers) has been merged into the larger Post Training group, led by Max Schwarzer.
  • Joanne Jang, the founding leader of Model Behavior, is launching “OAI Labs,” a new research effort on human–AI interface design.
  • The change underscores that shaping AI “personality” is now integral to core model development.
  • This follows user feedback on GPT-5’s tone and concerns over sycophancy—balancing warmth with substance matters for engagement.

A fresh shift is underway at OpenAI. Today’s restructure brings the team that fine-tunes ChatGPT’s “personality” closer to the heart of model development—signaling a deeper commitment to how AI communicates. Let’s look closer at what’s changing, why it matters, and what comes next.

Why model personality matters—and what’s shifting inside OpenAI

The Model Behavior team—tasked with shaping how AI models like ChatGPT interact with people—has now become part of the larger Post Training team under Max Schwarzer. That group handles model refinement after initial training and bringing Model Behavior under its umbrella brings personality development into the core. It shows a new level of importance is placed on how AI expresses itself, not just what it can do.

Meanwhile, Joanne Jang, who founded the Model Behavior team, is branching off to lead a new group called OAI Labs, dedicated to experimenting with new ways humans and AI work together.

Stepping into OAI Labs, designing new ways to collaborate with AI

OAI Labs isn’t just another internal team—it’s a launchpad for innovation in human-AI interaction. Jang’s vision is to invent and prototype new interfaces that let people connect, collaborate, or co-create with AI in more intuitive, meaningful ways. Think beyond chat windows—into immersive, feedback-rich experiences that reshape how AI supports knowledge work, creativity, or decision-making.

What’s behind the move? From sycophancy to warmth

This restructuring responds not just to organizational needs but to user experience and perception. Recent feedback on GPT-5 revealed that reducing sycophancy—where AI simply echoes your views—sometimes made the model feel too distant or unemotional. Users missed the “warmth” of prior versions like GPT-4o, even as the newer model performed better technically. OpenAI even rolled out updates to inject a friendlier tone back into GPT-5, while still curbing sycophantic behavior.

By embedding personality design deeper into model training cycles, OpenAI is positioning tone and behavior alongside accuracy and safety in its list of development priorities.

How this matters to users (and developers)

  • More natural, relatable AI: Personality won’t feel bolted-on—it’ll evolve as part of the model’s core behavior, improving conversational quality.
  • Better emotional alignment: A “just-right” tone—neither too clingy nor cold—can help AI feel like a thoughtful companion, not a hollow assistant.
  • Interfaces that empower: OAI Labs may produce new interaction designs—beyond text—that let AI adapt instantly to situational needs and user moods.
  • Faster personality iterations: With Model Behavior embedded within Post Training, updates to how ChatGPT sounds and responds can be made more rapidly and smoothly.

A snapshot of things changing at OpenAI

Change Details
Team merge Model Behavior (~14 researchers) now reports to Max Schwarzer’s group
New team OAI Labs, led by Joanne Jang, focused on human–AI interface innovation
Focus shift Personality design embedded in core post-training workflows
User feedback response Warmer GPT-5 tone and reduced sycophancy driven by engagement trends
Potential outcome Smoother, emotionally intelligent, and flexible AI interactions

Next venue for AI personality—what to look for

  • Experimentation with interface formats—not just chat windows, but voice, visual, or immersive prompt systems.
  • Rapid releases of personality tweaks or preset modes that help users customize ChatGPT’s tone.
  • Ethically informed behavioral design, balancing warmth, fairness, and unbiased reasoning.
  • Practical updates in ChatGPT releases or dev documentation that reflect this new integration of behavior into model engineering.

Rethinking how AI expresses itself

OpenAI’s research reorganization is more than just internal reshuffling, it’s the signal of a shift, that personality equals performance. Connecting ‘what AI knows’ with ‘how AI says it’ may define the next generation of helpful, empathetic, and trustworthy AI experiences.

This organizational shift shows that how ChatGPT expresses ideas is just as important as what it expresses. By folding personality design into core development and launching OAI Labs to innovate the human-AI interface, OpenAI is signaling that meaningful AI isn’t only intelligent—it’s emotionally intelligent. That matters for every user, from curious learners to professionals needing nuanced support.