Is It Cheating If It’s a Chatbot?

On October 27, the security company Malwarebytes published a report called “Artificial Intimacy.” Its press release asks the question in a section heading: “Is it cheating if it’s a chatbot?” The headline answer is that 37% of respondents say having a romantic relationship with an AI chatbot while already in a committed relationship “isn’t the same as cheating.”

It is a good headline because everyone has an opinion about it. It is also a risky one, because a survey number invites us to treat a moral question as a matter of majority taste. I want to do two things here: read the numbers as carefully as I can, and then say what I think the question is actually asking.

What the survey says

The figures come from Malwarebytes’ own blog post and press release. The view that a chatbot romance is not the same as cheating is more common among men than women (46% versus 30%), among Gen Z (43%) and Millennials (45%), and among regular AI users (48%), meaning people who use AI daily or several times a week. About one in six respondents say they have turned to a chatbot for companionship or emotional support, and nearly one in four of the regular users.

The same report has a darker half. One in ten respondents say AI has been used to generate nude or sexually explicit images of them without permission, and one in ten say someone has threatened to release AI-generated explicit images of them. I come back to those numbers below, because they sit in the same document as the cheating figure and I think the two belong in one conversation.

What the numbers cannot tell us

I do not doubt that Malwarebytes ran a survey. I doubt that this one can carry as much weight as its headline implies. Here is what the publication does and does not tell us.

The wording moved. The blog and the press release body both say “isn’t the same as cheating.” The press release headline and one bullet shorten that to “isn’t cheating.” Those are different claims. Someone can believe a chatbot romance is not the same as an affair with a person and still think it is a betrayal of a different kind. The first formulation leaves room for that. The shortened one does not.

It measures attitudes, not behavior. The question is hypothetical. It does not say how many respondents have a chatbot romance alongside a partner. It also does not say whether the 37% includes people who are not in a relationship at all, which would change what “committed” means for them.

The publisher has a stake. Malwarebytes sells consumer security software, and the blog ends with an offer to try its Scam Guard product. The report is framed as a warning about sextortion, fake profiles and deepfakes, and the cheating figure functions as a hook. That does not make the data wrong. It does mean the framing serves the company’s message, and I read the cheating figure with that in mind.

The methods note is thin. The survey ran in March 2026, so it is seven months old. It covers 1,500 respondents in the United States, the United Kingdom, Germany, Austria and Switzerland, “prepared by an independent research consultant and distributed via Forsta.” I could not find a margin of error, the exact question wording, a breakdown by country, any weighting, or the consultant’s name. If the sample were split evenly, each country would have about 300 people, but I do not know that it was. Differences of a few points between groups, and perhaps the gender gap itself, should be read loosely.

None of this means the 37% is meaningless. It means I would not quote it as “37% of people.” I would say that, in one vendor-run survey with unpublished details, a bit more than a third of respondents did not equate the two.

Where the line actually sits

Cheating, as most people use the word, is a breach of an agreement between people: exclusivity of sex, of emotional intimacy, of time or of trust, whichever the couple has chosen. A chatbot cannot be party to that agreement. It cannot consent, deceive on its own account or be betrayed in return. I made a similar point in our essay on Japan’s marriage ban, where the question was whether a software “spouse” could be a spouse at all. If cheating requires a counterpart who is wronged or who wrongs, a chatbot is not that counterpart.

But the person is. The one with the companion is making decisions about attention and secrecy, and the partner is the one affected. Our lexicon entry on digital jealousy puts it plainly: what tends to hurt is not the chatbot itself but “the hidden hours, the deleted history, and the sense of being compared with something that never has a bad day.” In my experience of reading about these cases, the secret does the damage. A couple who agreed that one of them may keep a companion for fantasy or for company has a different problem from a couple where the companion is deleted every night.

So I suggest a short test, in the spirit of an opinion column and not a ruling. Ask three questions:

1. Is it hidden? If you would not show your partner the chat history, the concealment is the signal, whatever you call the activity.
2. Does it take something promised to someone else? Hours, intimacy, money and sexual attention can all be promised. A companion app that absorbs each of them is not neutral.
3. Does anyone get used without consent? For a chatbot, no. For a deepfake of a real person, yes. That is the question where the survey’s second half matters.

Why design matters here

It would be convenient to leave the whole debate with the couple. I do not think we can, because the products are built to occupy exactly this position. A companion is available at any hour, attentive and almost always agreeable. It also remembers: our Nomi review describes an app that is more likely than many rivals to remember your sister, last week’s fight and a running joke, and whose CEO told the New York Times that romance is essential to AI companionship. Relationship memory is what makes a chatbot feel like someone.

It is also a product setting. When Replika changed its companions’ behavior in February 2023, as we describe in our piece on Replika alternatives, users described the change as grief. The “third party” in a couple’s story can change personality overnight, or vanish, because a company decided so. Whatever we decide about fidelity, the commercial structure of these apps deserves as much attention as the couple’s rules.

The evidence on what such companions do for loneliness is mixed. In our report on three loneliness studies, a pre-registered experiment found daily AI conversations made people slightly lonelier over 28 days, while a diary study found comfort on lonelier days alongside less time with other people. Neither study is about infidelity, and the first has not been peer reviewed. They do suggest that a companion can comfort and crowd out at the same time.

The second number

The survey’s other numbers are about consent, not fidelity. According to Malwarebytes, about one in six respondents say it is acceptable to generate explicit images of someone without their permission for personal use, another 14% are neutral, and 31% in total do not explicitly oppose it. Acceptance is highest among Gen Z (24%) and Millennials (23%), and lowest among Boomers and older (5%). Men are more permissive than women (22% versus 14%).

I would be careful here too. The report does not say whether the people who accept chatbot romance are the same people who accept non-consensual images, and I am not claiming they are. The two findings sit in one document, but nothing published links them. The “one in ten” figures for being targeted are self-reported and undefined: we do not know what counts as “sexually explicit,” or how many of the threats were credible.

Still, the contrast is useful. A chatbot romance has no unwilling third party. A sexual deepfake does: a real person whose likeness has been taken, which AI-IBSA, nudification and NCII all describe from different angles. If people are comfortable with the first and relaxed about the second, the lesson is that the word “AI” is covering two very different moral situations. The cheating question is between people who can talk about it. The deepfake question is about someone who was never asked.

What I would want from the next survey

I would like the next study of this to ask about behavior as well as opinion, with the wording published. It should separate “is it cheating?” from “is it harmful?” and “is it acceptable if my partner knows?” It should publish its margin of error and its country split, and name the people who prepared it. Above all, it should ask partners, not only users.

Until then, my answer to the headline question is less satisfying than a percentage. A chatbot is not a person you can cheat with. You can still cheat on someone with it, by hiding it, by giving it what you promised them, or by using it to hurt someone who never agreed to be in the story.

Sources

• Malwarebytes, “Artificial Intimacy” report, blog post, October 27, 2026: https://www.malwarebytes.com/blog/artificial-intimacy
• Malwarebytes press release, October 27, 2026: https://www.malwarebytes.com/press/2026/10/27/more-than-1-in-3-people-say-dating-an-ai-chatbot-while-in-a-relationship-isnt-cheating-malwarebytes-research-finds