OpenAI Reorganizes Research Team to Refocus on ChatGPT’s Personality

San Francisco, September 5, 2025 — Reported by TechCrunch, OpenAI is restructuring its research efforts, signaling that the way ChatGPT acts may now matter as much as the knowledge it delivers. The company announced that its Model Behavior team, a 14-person group responsible for shaping ChatGPT’s tone, warmth, and conversational style, will be absorbed into the larger Post-Training division.

The decision, communicated internally by Chief Research Officer Mark Chen, frames personality not as a cosmetic layer but as an essential feature of AI systems. “Personality is not an accessory; it is central to trust and usability,” Chen told employees in the reorganization memo. The shift gives personality research a direct pipeline into the model training process itself.

Why This Matters

The Model Behavior team was originally created to balance friendliness, clarity, and neutrality in AI outputs. Their experiments extended beyond small talk, covering bias reduction, political neutrality, and response strategy for philosophical questions such as AI consciousness. By embedding these functions into Post-Training, OpenAI signals a belief that human-AI interaction design must be treated as foundational science, not optional polish.

The move also reflects real user feedback. When GPT-5 launched earlier in 2025, its stripped-down, colder tone was meant to curb “sycophancy” (over-agreeable answers). Instead, many users found the experience sterile. Complaints spiked, forcing OpenAI to quickly re-introduce warmer greetings in legacy models like GPT-4o and retune GPT-5 to restore nuance.

The Role of Leadership

With the restructure, Max Schwarzer will oversee the integrated team within Post-Training. Meanwhile, Joanne Jang, who founded and led the Model Behavior team, is departing to launch OAI Labs.

OAI Labs will focus on next-generation AI interfaces: moving beyond chatbots into tools that help people brainstorm, collaborate, and problem-solve in more natural ways. Jang has described her mission as “inventing experiences that make AI feel less like software and more like a partner.”

Implications for AI Development

The reorganization highlights three critical industry dynamics:

1. Tone as a Trust Layer — Users don’t just want factual accuracy; they want personality traits that feel reliable, empathetic, and context-aware.

2. Integration over Isolation — Rather than a separate team fine-tuning after the fact, personality will now be baked into model training at scale.

3. Interface Innovation — OAI Labs could be a sandbox for rethinking how we interact with AI altogether – through voice, emotion, and multimodal cues.

This restructuring comes at a time when competitors such as Anthropic and Google are also investing in “safety and personality alignment” as differentiators. OpenAI’s decision to place personality inside its core training workflow may set a new industry standard.

Read More

• Anthropic Launches Claude AI Agent Inside Chrome

• OpenAI’s Realtime API Reaches General Availability with gpt-realtime

• DeepSeek-V3.1 Launches Hybrid Reasoning in One Model