OpenAI rolled out a memory upgrade to its ChatGPT app this week, and the feature does exactly what the name suggests: it now recalls details from your past conversations and weaves them into future ones.
Your job, your kids' names, the diet you mentioned in March, the trip you're planning for October.
It all gets filed away and quietly resurfaced.
The assistant feels less like a stranger and more like a coworker who actually listens.
For anyone juggling a dozen open loops in a single app, that continuity is genuinely useful.
But there's a reason this update is making people uneasy, and it has nothing to do with the technology working poorly.
Think about what most of us type into that little box.
We draft passive-aggressive texts we'll never send.
We ask about symptoms we haven't mentioned to our doctor.
We paste in financial details, relationship complaints, and half-formed business ideas.
Under the old model, each chat was a sealed room.
Now the rooms have doors, and the building remembers who walked through.
The opt-out exists, buried in settings, and you can clear stored memories.
But defaults shape behavior more than checkboxes do.
Most Americans will never open that menu.
They'll just notice, one day, that the app knows something they don't remember telling it — and feel a small chill they can't quite explain.
This is the part of the AI boom nobody puts in the keynote.
We are handing over the raw material of our inner lives, not because anyone forced us, but because the alternative is retyping context like a chump.
Convenience has always been the tax collector of privacy, and it never sends a receipt.
There's a second concern that's less discussed: memory drift.
A stored fact from six months ago may no longer be true.
The app doesn't know unless you tell it, and it may keep operating on stale assumptions with total confidence.
A confidently wrong one that thinks it knows you is worse.
None of this means you should delete the app.
It means the relationship changed this week, and most users weren't asked.
A tool that once answered questions is now building a profile.
That's a different product, whether or not the icon looks the same.
Parents should be paying closest attention.
Kids use these apps for homework, and they're far less likely to dig through settings menus.
A teenager's casual confession at 2 a.m. shouldn't become permanent context for a study session in June.
Our take: memory is the right feature shipped at the wrong moment, with the wrong defaults.
If OpenAI wants to be trusted with our past, it should ask first — clearly, once, in plain language — instead of tucking the choice three menus deep.
Final Thoughts
Until then, the smartest move is to check your settings and decide, deliberately, what this thing gets to keep.