OpenAI's terms of service for paid users may allow the company to use interaction data for model training, even if users have opted out. The distinction lies in whether "reasoning tokens" are considered "Output" protected by the opt-out policy. If these tokens, which contain processed user input and model responses, are not covered, OpenAI could potentially use them for synthetic data generation and model pretraining, undermining the effectiveness of the opt-out guarantee. AI
IMPACT Raises questions about user data privacy and the effectiveness of opt-out mechanisms in AI model training.
RANK_REASON Discussion of terms of service and potential implications for user data, rather than a direct announcement or release.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
