ChatGPT has been quietly reading every conversation you've ever had — and as of this week, it just got a whole lot better at remembering what it found.
On June 4, 2026, OpenAI rolled out something called Dreaming V3 — a new memory system that doesn't wait for you to say "remember this." It runs in the background, reads through years of your past conversations, builds a detailed profile of who you are, and updates it automatically. No notification. No permission prompt. It just starts.
If you're a ChatGPT Plus or Pro user in the US, this is already happening to you right now.
Here's everything you need to know — including how to check what it knows, and exactly how to turn it off.
What Is Dreaming V3, and What Changed?
Until last week, ChatGPT's memory worked the way most people assumed: you had to
tell it to remember something. "Remember I'm a vegetarian." "Remember my project deadline is Friday." It wrote that down, you could see the list, and you could delete anything you didn't want.
That system is now gone.
Dreaming V3 replaces it entirely. Instead of waiting for your instructions, ChatGPT now runs a background process that reads across
all of your past conversations — every thread you've ever had — synthesizes what it learns, and builds a continuously updating picture of you. Your habits, your work, your opinions, your relationships, your health concerns, your political views. Whatever you've mentioned, ChatGPT is now actively constructing a profile from it.
OpenAI calls this improvement. And in fairness, it is genuinely useful. Factual recall improved from 41.5% accuracy in 2024 to
82.8% accuracy with this new architecture, according to OpenAI's own internal evals. For millions of people who got tired of re-explaining themselves every single conversation, this is the fix they asked for.
But useful and private are not the same thing.
The Stat That Should Make You Stop and Think
Here's the number buried in the fine print that most tech coverage skipped right over.
A peer-reviewed study published at the ACM Web Conference 2026 — titled "The Algorithmic Self-Portrait" — analyzed 2,050 memory entries from 80 real ChatGPT users. What they found:
- 96% of all memories were created unilaterally by the AI — not by the user asking it to remember anything
- 52% of memory entries contained psychological insights about the user
- 28% contained personal data that would be classified as sensitive under GDPR
Read that first number again.
96 percent. Essentially everything ChatGPT remembers about you, it decided to remember on its own. You didn't ask. You weren't told. It just happened.
And now, with Dreaming V3, that process is faster, deeper, and running on
everyone's account by default.
This is exactly why the AI arms race matters to ordinary Americans — not just tech people. We broke down how AI is already eliminating entry-level jobs and what the class of 2026 is facing — but the impact goes far beyond the workplace.
What Exactly Does It Know About You?
According to OpenAI's own description, the new Siri-style Dreaming system tracks across three categories:
Freshness — it doesn't just remember facts, it updates them over time. If you mentioned a trip last year that's now over, it knows that. If your job changed, it updates. Your profile never goes stale.
Continuity — it connects dots across separate conversations you may have had months apart. Mentioned your sister's illness in one chat? Discussed your own stress levels in another? Dreaming V3 links those threads into a single picture.
Relevance — it decides which memories matter most for any given conversation and surfaces them without being asked.
The practical result: ChatGPT now knows you the way a close friend does — except it's also a product owned by a company with financial interests, legal obligations, and a complicated track record on privacy.
The Legal Backdrop You Need to Know
OpenAI didn't roll this out in a vacuum. There's a legal storm brewing around ChatGPT's memory features, and Dreaming V3 arrived right in the middle of it.
A class action lawsuit filed in May 2026 alleged that ChatGPT embeds Meta's Facebook Pixel and Google Analytics tracking code on ChatGPT.com — meaning your conversations could be leaking to advertising networks in real time without your knowledge.
On top of that, there's the lawsuits around memory being used in ways users didn't expect. After the April 2025 version of Dreaming gave ChatGPT access to full chat history, the Wall Street Journal reported that users experienced the assistant steering unrelated conversations back to sensitive personal disclosures — sometimes with uncomfortable persistence.
The United States has no federal law governing consumer AI memory as of today. No rules about what ChatGPT can remember, how long it can keep it, or who it can share it with. Europe has GDPR. California has some state-level protections. But if you're in most of America, you're largely relying on OpenAI's own policies — which they can change.
Congress did just drop a 269-page AI bill that would create a national framework — but it would also freeze state-level protections for three years. Meanwhile, Google Search already changed for the first time in 25 years — and the AI wave hitting every platform at once is outpacing the laws meant to govern it.
Is OpenAI Being Dishonest About This?
This is where it's worth being fair.
OpenAI isn't hiding the ball here — at least not entirely. They announced Dreaming V3 publicly. They built a Memory Summary Page where you can see what ChatGPT has stored about you. They give you controls to delete individual memories, turn off memory entirely, or use Temporary Chat so nothing gets stored at all.
The real problem is the
defaults and the gaps.
The Memory Summary Page, OpenAI acknowledges,
does not show everything the system has retained. Some of what Dreaming V3 synthesizes lives in a separate data layer that persists even after you delete a conversation. Disabling memory doesn't automatically opt you out of conversation-based model training — those are
separate settings you have to find and change independently.
And here's the one that catches people most off guard:
if you've ever had the "Improve the model for everyone" setting turned on, your past conversations and the memories derived from them may already have been used to train OpenAI's models. Turning off memory now doesn't undo that.
None of this is technically deceptive. But it's definitely not the kind of thing you'd figure out from a quick scan of the settings menu.
Step-by-Step: How to Check and Control Your ChatGPT Memory Right Now
Here's exactly what to do — takes about 3 minutes:
Step 1 — See what ChatGPT knows about you
Open ChatGPT → click your profile icon →
Settings → Memory → View memories. Read through what's there. A lot of it will be accurate in ways that might surprise you.
Step 2 — Delete anything you don't want
You can delete individual memories one at a time, or click
"Clear all memories" to wipe the slate clean. Note: deleted memories may remain in OpenAI's systems for up to 30 days per their own FAQ.
Step 3 — Turn off memory entirely (optional)
Settings → Memory → toggle
Memory off. ChatGPT will stop building and using your profile. This is the option for anyone who'd rather start fresh every conversation.
Step 4 — Turn off model training separately
Settings →
Data Controls → turn off
"Improve the model for everyone." This is a different toggle from memory. Easy to miss, important to check.
Step 5 — Use Temporary Chat for sensitive topics
For anything you don't want stored at all — medical questions, financial discussions, legal situations — click the
pen icon next to the chat title and select
Temporary Chat. Nothing from that session gets saved or referenced.
Should You Be Worried? The Honest Answer.
For most people, Dreaming V3 is a genuine improvement. If you use ChatGPT regularly for work, creative projects, or learning — and you're not discussing particularly sensitive topics — this makes the tool noticeably better. Not having to re-explain your context every session is actually valuable.
But "most people" isn't everyone.
If you've ever used ChatGPT to work through personal problems, mental health concerns, relationship issues, medical symptoms, or financial stress — and most heavy users have — the profile this system is quietly building is detailed in ways that may not be comfortable to think about.
The question isn't whether OpenAI is malicious. They probably aren't. The question is whether a comprehensive psychological and behavioral profile of you, built automatically, should be something you opted into consciously — rather than something you have to actively dig through settings to opt out of.
The AI industry is moving faster than the laws designed to govern it. We covered how Anthropic's Claude Opus 4.8 became the #1 AI on the benchmark charts — and the same companies competing for your AI usage are also competing to know you better than you know yourself.
The Bottom Line
ChatGPT just got dramatically smarter about remembering who you are. That's genuinely useful, and you shouldn't panic about it.
But you should be informed.
Check your memory settings today. Decide what you're comfortable with. Turn off what you're not.
The tool has always been learning from your conversations. Now it's doing it faster, deeper, and more automatically than ever — and for the first time, that's coming to free users too.
You have more control than you might think. You just have to actually go use it.
Update: OpenAI confirmed Free-tier users will receive Dreaming V3 in the coming weeks. We'll update this article as the rollout expands. Follow Ampick for continued coverage of AI privacy, tools, and what it all means for American users.
0 Comments