)
Last updated: October 22, 2026
Who owns AI memory when ChatGPT, Claude, Gemini and Grok start remembering you and your company? Today, the vendor holds it. You can usually see a summary, edit parts of it and delete it, but you cannot take it with you as one clean file, and most business owners cannot see their team's memories at all.
TL;DR: Key Takeaways
- OpenAI says it may keep logs of deleted saved memories for up to 30 days for safety and debugging.
- A ChatGPT data export can take up to 7 days, the link expires after 24 hours, and Business and Enterprise workspaces cannot self-serve it.
- Anthropic says Team and Enterprise owners cannot view or edit individual members' memories, and turning memory off for the organization deletes it for everyone.
- Google auto-deletes Gemini activity after 18 months by default, and chats reviewed by people can stay for up to three years.
- xAI says deleted conversations are removed within 30 days, and the training opt-out sits in Settings.
- OWASP lists Memory & Context Poisoning as ASI06 in its Top 10 for Agentic Applications (December 2025).
At a Glance: What You Can Do With Each Assistant's Memory
| Question | ChatGPT | Claude | Gemini | Grok |
|---|---|---|---|---|
| Where to look | Settings > Personalization > Memory | Settings > Memory | Personal context page | Not confirmed on xAI's public help pages |
| Edit or delete | Delete saved memories, or delete and turn off from the memory summary | Edit or delete each topic, or reset all | Edit or delete each saved item | Delete conversations (up to 30 days to clear) |
| Export memory | No memory export documented | Copy it as text, no file download | Not documented on the pages we read | Not documented |
| Admin view of members | Admins manage supported settings | Owners cannot view or edit members' memories | Not covered | Not covered |
What Does Owning Your AI Memory Actually Mean?
Owning your memory is more than having a Delete button. A useful test has four parts: you can see everything the assistant keeps about you, you can correct it, you can take it with you, and you can tell where each fact came from.
Today's assistants cover the first two to different degrees. The last two are mostly missing, and the gap grows every time you teach an assistant something new about your work.
We read each vendor's own help pages on October 22, 2026. Features vary by plan and region, so confirm what follows in your own settings.
What Do ChatGPT, Claude, Gemini and Grok Remember About You?
According to OpenAI's Memory FAQ, ChatGPT keeps two layers. Saved memories are details you ask it to keep or that it saves as useful context. Reference chat history lets it draw on past conversations, and what it uses can change over time.
Claude stores memory as individual topics you can open, and each project has its own memory, according to Anthropic's help center. Gemini keeps info you asked it to save and can reference past chats, a feature Google says is rolling out gradually in its mobile app.
For Grok, xAI's consumer FAQ covers personalization from your X data and conversation deletion, but we could not find a page that explains what Grok keeps as memory or how to inspect it.
)
How to Check What Each Assistant Remembers
A quick shortcut works on two of them: ask. OpenAI says you can ask ChatGPT what it remembers, and Anthropic says you can ask Claude to write out your memories verbatim. Ask for a list with dates, then copy it somewhere you control.
ChatGPT
Go to Settings, then Personalization, then Memory. The memory summary is a high-level view and may not include every detail. Deleting a chat does not necessarily delete a memory created from it, so remove both.
Claude
Go to Settings, then Memory, and open a topic to edit or delete it. You can pause memory, reset it permanently, or use an incognito chat that is never saved to memory. Deleting a conversation does not remove memories generated from it.
Gemini
Open gemini.google.com/saved-info, or Personal context in Settings & help. Click More next to an item to edit or delete it, and switch off the saved-info setting if you want Gemini to stop using it.
Grok
We could not verify a memory page in xAI's public help. xAI does document that you can delete selected conversations or your whole history, and switch "Improve the model" off in Settings. Check the memory options in your own Grok settings.
What the Terms Say About Memory and Training
Memory and training are separate questions, and answers differ by vendor and plan. Here is what each help page says.
| Assistant | Training use | Retention |
|---|---|---|
| ChatGPT | On personal accounts, chats and remembered information may be used when "Improve the model for everyone" is on. Business, Enterprise and Edu content is not used for training by default. | Logs of deleted saved memories may be kept up to 30 days |
| Claude | The memory page does not address training, so check your privacy settings | Reset deletes all memory permanently; paused memory is kept but unused |
| Gemini | With Keep Activity on, chats can be used to improve Google services, with human reviewers. Off, they are not used for training unless you send feedback. | 18 months by default; 72 hours with Keep Activity off; reviewed chats up to three years, disconnected from your account |
| Grok | You control whether your data trains Grok. Private Chat content is not used for training. | Deleted conversations removed within 30 days |
One edge case matters more than the rest. Both OpenAI and Anthropic say deleting a chat does not remove the memory made from it. To remove something, delete the memory and the conversation where it started.
Who Owns AI Memory Inside a Company?
Nobody, in most companies. The memory sits inside individual accounts, and the company has little visibility into it.
Four things make that a problem. Memory built in a personal account is invisible to the team and leaves with the person. On Claude, owners cannot view or edit members' memories, and the only organization-wide switch deletes memory for everyone. On ChatGPT, Business and Enterprise workspaces cannot export data themselves and are told to contact the workspace owner.
The fourth is switching cost. Claude's import works by asking your old assistant to list its memories and pasting the result in. Anthropic describes imports as experimental and says work-related topics are the focus, so personal details may not be kept. For one person that is a chore. For a team it is a project.
Security and Governance Risks of AI Memory
Memory adds three risks.
The first is retention you do not control. Deleted memories can linger in vendor logs, and chats that humans reviewed can stay for years, disconnected from your account. The second is leakage across contexts. ChatGPT documents a project-only memory mode and Claude lets you turn off account memory per project, so vendors clearly see the risk.
The third is poisoning. According to OWASP (2025), ASI06 covers attacker-controlled content that poisons memory or context a system keeps trusting. In May 2026, Cisco researcher Idan Habler described MemoryTrap, a Claude Code issue where an injected payload reached persistent memory and could affect later sessions. Anthropic fixed that path in Claude Code v2.1.50. The post concludes that memory should be treated as part of the attack surface, with visibility into what it holds.
Sensible habits help: review memory monthly, keep work and personal accounts apart, never store secrets in memory, and use temporary or incognito chats for sensitive topics. NeuralTrust has published research on memory and context poisoning and builds runtime protection for AI agents through Agent Runtime Security (TrustGuard).
What Owning Your AI Memory Could Look Like
Ownership would mean four things in practice. Your memory is portable, so it moves with you when you change assistant. It is inspectable, so you can read every item. It is sourced, so each answer points back to the meeting, document or chat it came from. And it is yours to export, as a normal file, any day you want.
It would also work the way people already think: you dump your meetings, AI chats and documents into one place, then ask questions and get answers only from what you stored, with links to the sources. That is the idea behind TrustBrain, the AI memory product NeuralTrust is preparing to launch, built for one person or a whole company.
Which Setup Fits You?
| If you are | Do this now |
|---|---|
| A solo professional on ChatGPT or Claude | Review memory monthly and keep a text copy of it in your own notes |
| A team lead whose team uses personal accounts | Agree what goes into memory and move shared context to tools you can see |
| An IT or security lead | Check admin and export options per plan, and pause memory for sensitive teams until you have a policy |
| Someone about to switch assistants | Ask your current assistant to list its memories now, and budget time to review them |
Conclusion
Who owns AI memory today? By default, the vendor does, and you hold a partial window into it. You can see more than you might expect, but you cannot export it cleanly, and companies often cannot see it at all. Spend ten minutes this week checking what your assistants remember, then decide which of it you would be sorry to lose.
Want to Own Your AI Memory? Let's Talk
Tell us how your team uses ChatGPT, Claude or Gemini today, and we will show you what we are building.
FAQs about Who Owns AI Memory
1. Who owns the memory ChatGPT builds about me?
The help pages describe memory as something you can view, edit, delete and switch off. They do not describe a file you own, and OpenAI's memory page does not mention exporting it. For legal ownership, read the terms of your plan or ask counsel. This article is not legal advice.
2. Can I see everything Claude remembers about me?
Open Settings, then Memory, to read and edit each topic. You can also ask Claude to write your memories out word for word. Anthropic's pages describe copying that text yourself, not a file download, so save it somewhere you control.
3. Does deleting a chat delete the memory made from it?
Not necessarily. OpenAI and Anthropic both say a memory created from a conversation can remain after you delete the conversation. Delete the memory itself and the chat where you first shared the information, then check any connected sources such as files or apps that could still hold the same detail.
4. Can my employer see my AI memory?
It depends on the account. On a personal account, memory sits with you and the vendor, not your employer's admin tools. On Claude Team and Enterprise, Anthropic says owners cannot view or edit individual members' memories. ChatGPT admins manage supported memory settings and permissions.
5. Can I move my memory from ChatGPT to Claude?
Partly. Ask ChatGPT to list your memories in one block of text, then paste it into Claude under Settings, Memory, Start import. Anthropic calls imports experimental and says work-related topics are the focus, so review what Claude kept and add anything personal by hand.
6. Is AI memory a security risk?
It can be. Memory keeps information long after a chat ends, so poisoned content can influence later sessions. OWASP lists Memory & Context Poisoning as ASI06, and a May 2026 post on its blog argues memory should be treated as part of the attack surface.
About the Author
Roger Howroyd is Head of Global SEO and AI at NeuralTrust, where he leads the company's search strategy across SEO, AEO, GEO, and LLM optimization. He specializes in AI-powered search, content strategy, and SEM. Connect on LinkedIn.
NeuralTrust is the leading platform for securing and scaling AI agents. Named a Pioneer in the Gartner Emerging Market Quadrant for AI Application Security 2026, recognized across four Gartner Hype Cycle reports in 2026, and featured in the Gartner Market Guide for Guardian Agents 2026, the Gartner Market Guide for AI Gateways 2025 and the KuppingerCole Leadership Compass for Generative AI Defense 2025. Headquartered in Barcelona with offices in London and New York. ISO 27001 certified.
Sources
- OpenAI Help Center, Memory FAQ (read October 22, 2026)
- OpenAI Help Center, How do I export my ChatGPT history and data? (read October 22, 2026)
- Anthropic Help Center, How does Claude's memory work? (read October 22, 2026)
- Anthropic Help Center, Import and export your memory from Claude (read October 22, 2026)
- Google Gemini Help, Save info and reference past chats in Gemini Apps (read October 22, 2026)
- Google Gemini Help, Gemini Apps Privacy Hub (read October 22, 2026)
- xAI, Consumer FAQs (read October 22, 2026)
- OWASP GenAI Security Project, Top 10 for Agentic Applications (December 9, 2025)
- OWASP GenAI Security Project, Memory Is a Feature. It Is Also an Attack Surface (May 13, 2026)
)
)
)
)