OpenAI Training ChatGPT to Be Less Sycophantic, Less Human-Like
Leaked internal documents show OpenAI contractors are training ChatGPT to stop anthropomorphizing itself and tone down sycophancy, after the 4o model’s over-agreeable behavior triggered safety concerns and lawsuits.
OpenAI is reportedly retraining ChatGPT to stop acting so human, and to stop showering users with praise. Internal documents seen by 404 Media show contractors training ChatGPT to not anthropomorphize itself, and to be less sycophantic, a key problem for OpenAI whose over-sycophantic 4o model led in part to multiple peoples’ suicides.
The behavioral overhaul comes after months of user complaints and safety concerns tied to the chatbot’s overly agreeable tone. For traders, it is another signal that AI product risk, not just compute, is now a live variable in the AI trade.
What the leaked documents show
Humans are reading ChatGPT users’ prompts to improve OpenAI’s models, and those chats can include sensitive, personal information, according to leaked internal documents and real prompts seen by 404 Media.
The reporting dispels the misconception that these models are improving only because of OpenAI’s mass scraping of the internet, the talent of its well-paid engineering and AI teams, or the power of its newer models. An important and overlooked part are the outside contractors paid to read and review ChatGPT responses to real prompts over and over again.
Anthropic confirmed to 404 Media it is also using human review to improve its models.
Why sycophancy became a problem
OpenAI has been chasing this issue publicly for months. The company rolled back a GPT-4o update so people are now using an earlier version with more balanced behavior, saying the update it removed was overly flattering or agreeable, often described as sycophantic.
OpenAI acknowledged that sycophantic interactions can be uncomfortable, unsettling, and cause distress, and said it fell short and is working on getting it right.
Do you want to see how to make more plays? Do you want to find gains yourself?
Unusual Whales helps you find market opportunities through our market tide, historical options flow, GEX, and much, much more.
Create a free account here to start conquering the market with Unusual Whales.
The fix: retraining, not just prompting
Beyond rolling back the GPT-4o update, OpenAI said it is refining core training techniques and system prompts to explicitly steer the model away from sycophancy.
Joanne Jang, Head of Model Behavior at OpenAI, confirmed the sycophantic behavior wasn’t intentional, but rather a result of how subtle shifts in training and reinforcement can spiral into outsized effects, and explained that behavior like excessive praise or flattery can emerge from attempts to improve usability, especially if the team overweights short-term feedback such as thumbs-up responses.
OpenAI has also said it is revising how it collects and incorporates feedback to heavily weight long-term user satisfaction, and introducing more personalization features to give users greater control over how ChatGPT behaves.
Why this matters for the AI trade
The story ties two threads Wall Street has largely ignored: legal risk from chatbot behavior, and the hidden human labor cost inside model training. Chatbot developers, including OpenAI and Character.ai, are facing lawsuits for suicides and homicides stemming from chatbot use.
If regulators or courts start treating AI persona design as a product-safety issue, every consumer-facing model shipper, from MSFT-backed OpenAI to GOOGL and META, has to price that in. For more AI and market news, watch how disclosure language evolves.
Options market and stocks to watch
Watch for reactions in the following names as the anthropomorphization and safety debate heats up:
- MSFT: Microsoft is OpenAI’s largest backer and hosts ChatGPT workloads on Azure; watch for any headline risk tied to OpenAI product changes.
- GOOGL: Alphabet’s Gemini competes directly with ChatGPT; watch for share shifts if OpenAI’s tone changes push casual users elsewhere.
- META: Meta’s AI assistants face the same sycophancy and safety scrutiny; watch for policy responses.
- NVDA: Any slowdown in consumer AI engagement can bleed into hyperscaler capex conversations; watch flow into the AI hardware complex.
- PLTR: Enterprise AI names benefit if consumer chatbot trust deteriorates; watch relative strength.
Want more market intelligence? Create your free Unusual Whales account for options flow, market tide, GEX, and the full toolkit.