ChatGPT's new memory upgrade is powerful - and could poison every answer it gives you
Summary
ChatGPT's new Dreaming V3 memory upgrade, introduced in 2026, significantly expands how the AI retains user context by synthesizing a personal profile from past interactions, including chat history, explicit instructions, and implicit preferences. OpenAI reports V3 improved factual task recall from 41% to 82%, correctness over time from 9% to 75%, and preference adherence from 31% to 71%, while reducing compute costs by 5X. However, the author raises concerns that these automatically curated memories are often inaccurate, retain outdated information, and can subtly skew future AI responses by filtering information through a potentially flawed personal lens. This feature is rolling out to all users, currently available for Plus and Pro subscribers.
Key takeaway
For users relying on ChatGPT for diverse tasks, be aware that its new Dreaming V3 memory system actively synthesizes a personal profile from your chat history. This profile can be inaccurate, retain outdated information, and subtly bias future responses. You should critically evaluate all AI outputs, as disabling memory or deleting specific items may not fully purge the AI's internal representation of you, potentially leading to skewed or incomplete information.
Key insights
ChatGPT's advanced memory system, while technically efficient, creates an often inaccurate and unmanageable user profile that can bias responses.
Principles
- AI memory systems can synthesize user profiles.
- Automated memory curation may retain irrelevant data.
- User controls for AI memory may be incomplete.
Method
Users can access memory settings via Settings > Personalization > Memory to disable memory (partially), delete saved memories, or manage the consolidated profile, though full data purging requires deleting chat sessions.
In practice
- Review ChatGPT's generated memory profile.
- Delete specific chat sessions to purge data.
- Disabling memory doesn't remove all stored data.
Topics
- ChatGPT
- AI Memory
- User Profiling
- AI Bias
- Data Privacy
- Large Language Models
Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Product Manager, AI Ethicist, General Interest
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by News and Advice on the World's Latest Innovations | ZDNET.