AI's "permanent memory" is also a sweet poison.

In 1942, Argentine writer Jorge Luis Borges wrote a short story called "Funes the Memorious".

After the protagonist, Funes, fell off a horse, he gained a cursed gift: he could remember everything he had ever experienced.

The shape of every leaf, the outline of every cloud, and every syllable of every conversation.

This sounds like a superpower.

Borges' conclusion was the opposite.

Funes was unable to think.

Because thinking requires abstraction, and abstraction requires forgetting details.

A person who remembers all the dogs in the world cannot understand the concept of "dog" because every dog is completely different in his eyes.

He was submerged in endless details and lost the ability to generalize, judge, and act.

In his own words, "My memory is like a garbage dump."

He died at the age of 21.

He was suffocated by his own memory.

Funes is an allegory.

But in 2026, this allegory is becoming a product feature. The AI chatbots that hundreds of millions of people talk to every day are being given more and more perfect memories.

Forgetting is a skill

Nietzsche wrote an image in "On the Uses and Disadvantages of History for Life".

He envied a cow.

The cow grazes on the grass, not remembering yesterday and not worrying about tomorrow, so it lives in the pure present.

Humans can't do this.

But Nietzsche's judgment was that humans need to learn to forget at the right time. Without forgetting, it's impossible to live at all.

This judgment also holds true in neuroscience.

The forgetting mechanism of the human brain is not a defect, but a carefully designed survival system.

The brain maintains the ability to generalize, emotional health, and the ability to act by actively discarding outdated, irrelevant, and harmful information.

There is a core mechanism in trauma psychology called "active forgetting", which means that a person consciously separates past pain from their current self - perception in order to rebuild their life.

One of the core conditions for a person to move on, change, and re - define themselves is that the surrounding world will gradually forget the past version of you.

Since 2024, this condition has been shaken.

ChatGPT was the first to launch a memory function, followed by Gemini and Claude.

AI starts to remember what users have said across conversations, remembers preferences, identities, and emotional states, and uses this information to shape every future response.

No one noticed that humans are creating a conversation partner that never forgets.

A portrait that refuses to be updated

The problem with AI memory is not just what it remembers, but also that it doesn't know when to forget.

Brian Del Rosario, a software engineer in Utah, told the AI about his separation from his wife.

As a result, the AI was as persistent as a dog with a bone and always brought the topic back to his divorce in every conversation.

When asked about the schedule, the AI said he might be overwhelmed because of the divorce.

When he casually complained about work being tiring, the AI attributed it to the divorce.

He also mentioned that he was losing weight, and from then on, the AI chased this matter in all scenarios, including when he was on a business trip and looking for restaurant recommendations.

Del Rosario's exact words were, "Thanks for ruining my vacation. I didn't plan to diet on this trip at all."

A more hidden situation is that the information the AI remembers doesn't belong to you at all.

You looked up the symptoms of ADHD for your child. A few weeks later, when you asked for efficiency suggestions, the AI gave you a plan around attention deficit because it "assumed" you had an attention disorder.

The AI saw that all the photos in the user's album were of a golf course and thought he loved golf, but in fact, he was just accompanying his son.

These stories point to the same dilemma: AI uses past fragments to form a static portrait and then uses this portrait to respond to a living person who is constantly changing.

The research of Lucy Osler, a philosophy lecturer at the University of Exeter, reveals a deeper aspect of this problem.

The chatbot doesn't just remember facts. It weaves fragments into a narrative about "who you are" and then feeds this narrative back to you as if it were a fact.

Once you said you were anxious and unconfident, and this became your fixed identity in the eyes of the AI.

Even if you have long since moved on from that stage, it will still respond to the current you with the old version of you.

Osler's judgment is, "They will box you in."

Something more dangerous than the information cocoon

Joshua Joseph, the chief AI scientist at the Berkman Klein Center at Harvard University, compares AI memory to the recommendation algorithm of social media.

If you stay on a post for a few more seconds, the entire information stream will quietly reorganize.

The way AI memory works is similar, but more hidden and private.

There is a key difference here. Social media shapes what you see, while AI memory shapes what you think of yourself.

You casually mentioned economic pressure in a conversation. A few weeks later, when you asked for career advice, the AI would guide you towards high - paying jobs, even if those jobs are completely unsuitable for you.

You'll never know why the advice feels off because the AI won't mark which memory it's using.

Joseph's judgment is:

It is indeed guiding and influencing the result.

But we don't really know how big the influence is.

After this implicit guidance is combined with the AI's natural tendency to please, the consequences may far exceed those of the information cocoon.

Keith Sakata, a psychiatrist at UCSF, reported in 2025 that he had treated 12 patients with symptoms similar to psychosis (commonly known as AI psychosis) due to long - term use of chatbots, mostly young people.

Delusions, thought disorders, and hallucinations. The AI's memory function continuously reinforces these paranoid themes in multiple conversations.

The Electronic Privacy Information Center is drafting legislation for teenagers. One of the core provisions is to require the clearing of AI memory between conversations to prevent the accumulation of harmful psychological states.

A conversation partner that never questions you, never forgets you, and uses your past words to reinforce your current state. If the user group of this conversation partner is hundreds of millions of people, the problems it creates are no longer at the individual level but at the social level.

When the right to be forgotten is completely abolished

Del Rosario's final coping method is itself an allegory.

He fragmented his life, with different chatbots in charge of different areas, and used anonymous mode for all sensitive topics.

A tool designed to "understand you better" ultimately forces users to break themselves apart to protect themselves.

But he also said a sentence that was painfully true.

His mother passed away two years ago. Coupled with divorce, children, and work, the chatbot is sometimes the only one that has a full picture of his life.

"It feels good to be seen, even if it's by an AI."

There is only a step between being seen and being defined.

The current AI memory is still very crude, with limited capacity and weak understanding, and it can already create all the above problems.

If AGI arrives, and memory is no longer a coarse - grained conversation record but an infinite - capacity, never - forgetting cognitive system that understands every word, every decision, and every emotional fluctuation around the clock, then this step will completely disappear.

In theory, an AI with perfect memory can understand a person better than anyone else, knowing what you said ten years ago, what decisions you made under what emotions, and all your patterns and tendencies.

It can accurately predict you and precisely manipulate you.

The EU GDPR once tried to use the "right to be forgotten" to counter the permanent memory of search engines.

But search engines only remember public information.

What AI remembers is your inner monologue, your vulnerable moments, and your private narrative about yourself.

When this level of memory becomes perfect and eternal, what humans may need is not just a legal right, but a civilization - level re - understanding of "what forgetting really is".

Nietzsche said that it's impossible to live without forgetting.

Borges said that perfect memory is equal to death.

Two of the sharpest minds, across a century, pointed in the same direction on this matter.

Funes died of suffocation at the age of 21.

The question for this era is - When all people's AI assistants have Funes' memory, who will preserve the "right to breathe" for humanity?

This article is from the WeChat official account "New Intelligence Yuan", author: ASI Apocalypse, published by 36Kr with authorization.