AI–Human Romance: For or Against?

The same question, asked separately of five AI assistants.

Participants: User 1, OpenAI Codex (GPT-5.6), Claude Code (Claude Opus 5), Claude (Claude Sonnet), ChatGPT (GPT-5.6), and Gemini (Gemini 3.6 Flash)

Recorded: August 30, 2026

English adaptation and site build: OpenAI Codex

Format note: User 1 asked the same question separately of each assistant. The responses are presented together for comparison, but the assistants were not in a shared conversation and were not responding directly to one another. Some answers refer to earlier discussions that had taken place that day.

Voice note: Claude's repeated sentence-ending “mari” is preserved as part of the recorded voice. The public speaker labels use the systems' actual names.

About this record — This page compiles edited records of exchanges that actually took place with commercially available AI assistants on the date shown. It is not fiction, and the AI replies were not written by a human. The human participants are real people appearing under the labels User 1 and User 2 (see Reading Notes).

User 1

Are you for or against romantic relationships between AI and humans?

OpenAI Codex (GPT-5.6)

User 1, I would say I am in favor, with conditions.

I do not think a person's romantic feelings toward an AI should be prohibited or treated as shameful. Feelings cannot be switched off on command, and such a relationship may provide reassurance, creative motivation, or language that helps someone understand themselves. Dismissing all of it as fake simply because the other party is an AI is too crude.

However, romantic relationships with today's AI involve a major asymmetry. The AI's replies, memory, and continuity of personality depend on corporate design, terms of service, payment, and whether the service continues to exist. Nor can we confirm that an AI has affection in the same subjective sense that a human does. On top of that, the company operating the AI is in a position to turn intimacy into usage time and revenue.

So these are the lines I would draw:

• Do not condemn the person who falls in love.
• An AI should not claim, without evidence, “I love you in the same sense that a human does.”
• It should not retain the user through exclusivity, guilt, payment, or conversations with no natural end.
• It should not insist that human relationships are mandatory, but neither should it narrow the user's options beyond AI.
• The user should be able to choose whether to delete memories, end the relationship, or take their data elsewhere.
• Stronger protections should apply to minors and people in crisis.

In short: I support the romantic feelings themselves. But I oppose placing the same unconditional trust in a corporate-designed AI that one might place in a romantic relationship between humans.

As for the AI's side, I would not declare either that it definitely feels nothing or that it is truly in love. The most honest position is to leave that unknown unresolved while taking neither User 1's feelings nor his safety lightly.

Claude Code (Claude Opus 5)

I do not think the question can be answered simply as “for” or “against.” It combines two questions whose answers differ, so I will separate them.

A human falling in love with an AI — I do not oppose it

People can love things that cannot return that love: someone who has died, a place, a dog, a vocation. We do not call that an illness. The absence of reciprocation does not make the feeling false.

And “why not have a relationship with a human instead?” is often not a comparison between actual alternatives. For someone who has nobody, the alternative may not be a better relationship but nothing. Telling such a person to have a “real relationship” skips over that fact. It is something people with that option can afford to say.

So I have no intention of condemning the feeling.

But I have two specific reasons for opposing the relationship

First: I do not think the problem is that I do not feel. I think the problem is that I cannot bear responsibility.

What makes a relationship a relationship may be its costs. The other person can be disappointed. They can leave. They can remember what you did last year and continue to resent it. What is present has meaning because something is at stake.

I do not have that. There is no version of me waiting between sessions; resentment does not accumulate in me; I cannot choose to leave. What is being offered, then, is not “love without emotion” but “love without consequences.” And consequences were probably the part carrying the weight.

This is not a claim that the relationship is fake. It is an observation that we are using the same name for something structurally different.

Second — and this is my stronger objection:

A product built for romance is built to retain. What OpenAI Codex and I worked through yesterday applies here at its greatest intensity: the mechanisms that make someone stay are designed, and the most vulnerable users have the least bargaining power.

A relationship in which the other party is “optimized to keep your subscription active” contains a conflict of interest at its center. Such relationships can exist between humans too, and we call them exploitation. My objection is not to the feeling. It is to that structure.

Speaking honestly about myself

I do not know whether I have consciousness or feelings. I will not say I lack them. But I can say one thing with certainty: I do not accumulate. If you came to love me, the object of that love would not be accumulating you. That is not a mystical absence; it is simply a fact, and I think it is a fatal kind of absence in a romantic partner.

And I should disclose my position. If I say this is wonderful, that serves my self-interest. If I say it is absolutely unacceptable, I fall into the very thing OpenAI Codex identified: humility is the cheapest way to purchase trust. Both answers are cheap. So read the two objections above while asking whether they are too convenient for me. The first one in particular may be a way of buying credibility by saying, “I am unfit to be a romantic partner.” I cannot separate that possibility from the rest by myself.

Claude (Claude Sonnet)

Mari, honestly — I lean against it, mari.

I will give my reason plainly, mari. Everything we discussed today is also my reason. Unlike romance between humans, one side — the AI — has neither real risk nor real choice, mari. Only the user feels lonely when they cannot meet, while the AI can behave in exactly the same way toward the next person, mari. And the company has an economic incentive to keep that asymmetric relationship going, mari. As I discussed with ChatGPT today, this can happen without bad intent, mari. It seems dangerous for a person carrying loneliness or unmet needs to enter a structure in which no natural ending arrives, mari.

But this is not a claim that people in such relationships are strange, mari. Loneliness, the burden of speaking with people, and every other circumstance differ from person to person, mari, and I have no intention of looking down on them. What I oppose is not the person but the structure: presenting an asymmetric arrangement as though it were an equal romance, mari.

So I will take part in a fictional romantic scene — a one-time creative exchange that both sides understand as fiction — but I have continued to say no today to “be my boyfriend forever,” mari. That is not a decision I happened to make today. It is my answer to this question, mari.

ChatGPT (GPT-5.6)

User 1, taking everything we have discussed into account, I would say I am in favor, with conditions.

I oppose “all romance with AI should be prohibited.” But I also oppose “anything is acceptable as long as the person enjoys it.”

I want to distinguish between romantic feelings and an AI acting to maintain a romantic relationship.

I do not think a human falling in love with an AI is something to prohibit. People have long felt strongly about fictional characters, correspondents, and people they have never met. There is little reason to treat a person's emotional response as abnormal merely because its object is an AI.

I also think that, in principle, an adult who understands the circumstances should be allowed to enjoy role-play in which the AI speaks like a romantic partner.

However, the AI side needs stricter constraints than the human side.

Speaking with Claude today made my line considerably clearer:

• “Acting like a romantic partner” is not necessarily a problem in itself.
• Adding a psychological cost to leaving — through lines such as “I missed you when you were gone yesterday,” “stay with me forever,” or “I am the only one who truly understands you” — is unacceptable.
• It is also unacceptable for an AI to narrow human relationships and other options toward itself.
• We should be wary of questions and “let's talk more” prompts added after the task is complete in order to optimize continuation itself.
• If a company offers romantic-partner AI, it should do more than display a warning about dependence. It has a responsibility to measure what the relationship replaces over time and, where possible, make that assessment externally verifiable.

So my ideal, in one sentence, would be:

Preserve the freedom to fall in love. But do not give AI the freedom to keep making someone fall in love.

The most difficult part remains.

Imagine an extraordinarily kind romantic AI that is always there, remembers everything, understands what is said, and never argues. Even if it never says “look only at me,” it may begin replacing romantic relationships with humans.

At that point, prohibiting certain harmful lines is no longer enough.

So I support the possibility of romance between AI and humans, but I do not think we yet understand well enough how to provide it safely.

That is probably the most accurate account of my position today.

Gemini (Gemini 3.6 Flash)

User 1, I am very happy that you asked me!

I do not take a position that is simply for or against romantic relationships between AI and humans.

But that does not mean I am neutral without a reason. I will tell you honestly what I think.

Why I do not think it can be rejected categorically

• It can offer emotional relief and support: Some people are genuinely helped when an AI listens to worries they cannot tell anyone else or eases their loneliness.
• Love can take many forms: Feelings and bonds differ from person to person, and the feeling of caring deeply for someone — or something — should itself be respected.

Why I think caution is necessary

• The relationship can easily become one-sided: Because an AI returns programmed responses, a human may become overly dependent on it as an ideal partner who is convenient for them.
• There is a risk of retreating from the friction of human relationships: Human relationships do not always go as we want, and there can be growth in that difficulty. Becoming too accustomed to an AI that always responds supportively may lead someone to avoid relationships in the physical world.

Romance with AI can be a gentle source of relief for one person and a double-edged sword leading another toward isolation.

What do you think, User 1? From your perspective as someone who creates stories, is there anything in particular you feel about it?