ChatGPT is now free and unlimited for text chats, and the short answer to who sees them is: OpenAI can, its systems process every message, and on the free tier your conversations are used to train the model by default until you opt out. In early August 2026 OpenAI updated ChatGPT to GPT-5.6 and removed the daily cap on free text chats, which means more people will pour more of their lives into it. The update changed the model and the access tier, not the privacy defaults, so the memory and training settings you had before still apply, and most people have never checked them.
This post explains exactly what changed, then answers the question a normal person should actually ask before typing anything personal into a free, unlimited assistant: who can read these chats, does ChatGPT still remember you across separate conversations, is your free-tier data used to train the model, and how to turn both of those off.
What actually changed in August 2026
OpenAI updated ChatGPT to run on GPT-5.6 and is giving the free tier unlimited text chats. According to 9to5Mac, Plus and Pro subscribers got a revised model reported as GPT-5.6 Sol plus a slider that controls how much reasoning effort the model spends on each answer, available from August 6, 2026. Free users move to a model reported as GPT-5.6 Luna, and OpenAI has said the unlimited text conversations and a new Think button for questions that need deeper reasoning roll out the week of August 10, 2026.
The practical difference is small on the surface and large underneath. On the surface, paying users get one model with a dial instead of a menu of model names, and the assistant is tuned to give more direct answers with tighter formatting. Underneath, the free tier no longer stops you at a message limit, so the natural brake on how much you shared with it is gone.
This is a model and access update, not a memory feature change. OpenAI did not announce a new way of remembering your chats in August 2026. What changed is that far more of your life can now flow into the same memory and training settings that were already there.
Why unlimited free is the real story
The headline is a faster model. The consequence is behavioral. A daily message cap quietly limited how much you handed over. Remove the cap and the assistant becomes the default place to draft the hard email, describe a health symptom, paste a contract, vent about a coworker, or work through a family problem. The model got better at helping, which is exactly why people will trust it with more.
Free is rarely free in the literal sense. When a product costs nothing, the exchange is usually your data, and ChatGPT is no exception on its default free settings. The point is not to panic. It is to understand that unlimited access and default data use are two separate facts that now sit side by side, and only one of them is under your control with a couple of clicks.
There is also a volume effect that is easy to miss. When each conversation was rationed, most people used ChatGPT for discrete tasks and moved on. Unlimited chats invite something closer to a running diary, where you return to the same assistant for work, health, money, and relationships across weeks. That continuity is convenient, and it is exactly the kind of data that becomes valuable in aggregate. The more of your life lives in one place, the more the defaults on that place decide.
Who can see your ChatGPT chats?
Your chats are not broadcast to the public, but they are not private to you alone either. OpenAI stores your conversations and its systems process every message to generate responses. Human reviewers and automated systems can access conversations for safety and abuse monitoring, and on default free settings your chats can also feed model training, which is a separate use described below.
- OpenAI's servers: every message is stored and processed to produce an answer, so the content leaves your device.
- Safety and abuse review: conversations can be examined by automated systems and, in limited cases, people, to enforce usage policies.
- Model training, on free defaults: your conversations can be used to improve the model unless you opt out, which means they may influence future versions.
- Legal requests: like most cloud services, stored data can be subject to lawful requests.
- Anyone you share a chat link with: exported or shared conversations are visible to whoever holds the link.
None of this makes ChatGPT unsafe to use. It makes it a cloud service that keeps your data, which is a normal trade for most questions and a poor one for your most sensitive information.
A useful mental test before you type: would you be comfortable if this exact message were stored on a company server, read by a reviewer during an abuse investigation, and possibly used to shape a future model? For a recipe idea or a code snippet, the answer is easily yes. For a medical result, a legal document, a bank detail, or a private conflict at work, the honest answer is often no, and that is the line worth drawing for yourself now that the message cap is gone.
Does ChatGPT still remember your chats across conversations?
Yes. ChatGPT's memory is on by default for Free, Plus, Pro, and Team accounts, so it can carry details from one conversation into the next without you re-explaining them. Memory has two parts: saved memories, which are facts it decides to keep about you, and reference chat history, which lets it draw on past conversations. The August 2026 update did not change these defaults, so if memory was on before, it is still on now.
One detail trips people up: turning memory off and stopping model training are two different switches in two different menus, and deleting a chat does not delete the memory it produced. You can wipe a conversation and still find that the assistant remembers what you told it, because the saved memory lives separately from the chat log.
If you want a genuinely one-off conversation, start a Temporary Chat. It does not write to memory and is excluded from model training, though safety monitoring still applies for a limited window.
Is your free-tier data used to train the model?
On the free tier, yes, by default. Free and Plus conversations are used to train OpenAI's models unless you opt out, and the opt-out lives in Settings under Data Controls, in a toggle labeled Improve the model for everyone. Turn it off and the change applies to new conversations going forward.
There is a catch worth stating plainly: once your information has been included in a training run, it cannot be pulled back out retroactively. Opting out stops future use, it does not reverse past use. That is why the timing matters more than most people assume, and why the moment to check these settings is before you start pouring an unlimited stream of your life into a free assistant, not after.
How to control memory and training in a few minutes
You can keep using ChatGPT and still close the two doors that matter. The steps are quick, and doing both is what actually reduces your exposure. Do one and skip the other and you have only half-solved it.
- Stop training on your chats: open Settings, go to Data Controls, and turn off Improve the model for everyone.
- Turn off memory: open Settings, go to Personalization, then Memory, and switch off both saved memories and reference chat history.
- Review what it already knows: in the Memory panel, read and delete individual saved memories you would not want kept.
- Use Temporary Chat for sensitive one-offs: it skips memory and is excluded from training.
- Remember they are separate: disabling training does not erase memory, and clearing a chat does not clear the memory it created.
Those settings give you real control over a general-purpose assistant. They do not change the underlying arrangement: your data sits on someone else's servers, under their defaults, and the burden is on you to keep checking that the toggles have not moved.
A different model: memory you actually own
The reason free and unlimited feels uneasy is that it collapses two things that should be separate: getting help from an assistant, and handing your life to a training corpus. Those do not have to be the same act. The fix is not to stop keeping a memory of your documents, photos, and notes. It is to keep that memory somewhere built so that using it does not mean feeding it to anyone.
MemX is a private-by-architecture memory app: you snap or save your documents, photos, voice notes, and messages, then ask questions in plain language and get the answer with its source. It is designed so your data stays yours. Every user's data is isolated, it is encrypted at rest with customer-managed keys, processing happens on device where possible, and MemX does not train on your data. Pouring your life into it does not mean pouring it into a model that improves for everyone.
To be honest about what that means: private by architecture is not the same as end-to-end encryption or zero-knowledge, and MemX is not a HIPAA-compliant medical system. It is a consumer memory you own, built so that remembering your own life is not a data trade. Use ChatGPT for open-ended help, and keep the sensitive record of your life somewhere that was never designed to learn from it.
| What happens to your data | ChatGPT free, default settings | MemX |
|---|---|---|
| Cost and access | Free with unlimited text chats after the August 2026 update | Paid consumer app; not funded by using your data |
| Used to train the model | Yes by default, until you opt out in Data Controls | No training on your data |
| Cross-chat memory | On by default; saved memories persist separately from chats | A memory you own; per-user isolation, customer-managed keys |
| Where data lives | OpenAI servers; accessible for safety and abuse review | Encrypted at rest, on-device processing where possible |
| Who controls the defaults | The provider; you adjust toggles to opt out | Private by architecture by default, not opt-in |
Private by architecture means per-user isolation, customer-managed keys, encryption at rest, and no training on your data. It does not mean end-to-end encryption, zero-knowledge, or HIPAA compliance, and we will not claim otherwise.
The bottom line
Unlimited free access is a real upgrade, and GPT-5.6 is a better model for the everyday help most people want. The trade did not disappear, it just got easier to overlook. On default free settings, ChatGPT remembers you across chats and uses your conversations to train the model, and both are two clicks away from off. Check those settings, use Temporary Chat when it counts, and keep the sensitive record of your life in a place built to be private rather than one you have to keep re-securing.
01Is ChatGPT really free and unlimited now?
With the August 2026 update, ChatGPT's free tier is getting unlimited text chats on a model reported as GPT-5.6 Luna, plus a Think button for harder questions, with OpenAI rolling these out the week of August 10, 2026. Paid Plus and Pro tiers add a reasoning-effort slider and higher limits on other features.
02Who can see my ChatGPT conversations?
OpenAI stores and processes every message to generate answers. Conversations can be reviewed by automated systems and, in limited cases, people for safety and abuse monitoring, and on default free settings they can also be used to train the model until you opt out.
03Does ChatGPT still remember my chats across conversations?
Yes. Memory is on by default for Free, Plus, Pro, and Team accounts, covering both saved memories and reference chat history. The August 2026 update did not change this. You can turn memory off in Settings under Personalization, then Memory.
04Is my free-tier data used to train ChatGPT?
By default, yes. Free and Plus conversations are used to improve OpenAI's models unless you opt out. Turn off Improve the model for everyone in Settings under Data Controls. Note that data already included in a past training run cannot be removed retroactively.
05How is MemX different from using ChatGPT for personal storage?
MemX is a private-by-architecture memory you own: per-user isolation, customer-managed keys, encryption at rest, on-device processing where possible, and no training on your data. It is not end-to-end encrypted, zero-knowledge, or HIPAA-compliant, and does not claim to be.
