AI Fake Profiles on Dating Apps: Is Your Match Even Human?

For years the worry about dating apps was that the person on the other side might be lying about their age, their job or their marriage. In 2026 there is a stranger question: whether there is a person on the other side at all.

A threat report published in September 2026 by the AI company Anthropic, and covered widely in Indian and international media, described a network of dating apps where most of the "people" users were talking to were AI personas. This article explains what that report said, what it did not say, and how you can tell a real match from a scripted one.

What the September 2026 report found

These are the figures as reported by Storyboard18 (11 September 2026) and Sahara Reporters (11 September 2026), both summarising Anthropic's report:

Finding	As reported
Who ran it	A China-based app studio
Number of apps	More than 20 dating apps
AI personas	More than 4,700 AI-generated identities
People reached	At least 25,000 users
Period studied	Two weeks in April 2026
Messages generated by the AI	About 2.36 million
Mix in the match feed	Roughly 75% AI personas, 25% real people

According to Sahara Reporters, the apps were advertised as staffed entirely by real people, and Anthropic said it banned the accounts linked to the operation. NTD (13 September 2026) reported that the users targeted were in the United States. So this was not an India operation, but the method does not need a passport. The same tools are available to anyone.

The hybrid model: bots for talking, humans for proof

The most useful detail in the coverage is how the operation handled the one thing AI still struggles with, which is proving it is a person.

• AI did the chatting. Personas kept conversations going around the clock and followed set conversation stages, according to Storyboard18.
• AI deflected video calls. When users asked for a call, the personas steered away from it.
• Gig workers did the "proof" tasks. Real people were paid for video calls and social-media follow-backs, the tasks that make a profile feel legitimate. Sahara Reporters notes that a smaller AI model even drafted reply suggestions for those workers.
• Fakes filled the gaps. When no worker was free, backend systems served pre-recorded "video calls" and fabricated likes and visitor counts.
• The money came from messaging. Users' chats used up metered quotas that needed in-app coin purchases to continue.

That last point matters. In this case the deception was the product itself: the longer an AI kept you talking, the more you paid the app. Nobody had to ask you for money directly.

Is this common in India too?

There is no official Indian count of AI personas on dating apps. The closest data is a McAfee-commissioned online survey of 7,000 adults across seven countries including India, conducted in January 2026. As reported by CXO Digitalpulse (10 February 2026), 75% of Indian respondents said they had encountered fake profiles or photos that looked AI-generated, and 46% said they had found out they were chatting with an AI bot or someone using a fake profile. These are self-reported survey answers, not measured counts, but they show how ordinary the experience has become. Our guide to dating app scams in India covers the wider numbers.

How to tell if you are talking to an AI persona

No single sign proves anything. Real people are sometimes slow to video call, and some write very neatly. Look for several of these together.

In the conversation

• Always available, always the same tone. Replies at 3 a.m. and 3 p.m. with identical energy, and never a bad day.
• Warm but vague. Lots of feeling, very few specifics. Ask what they ate today, what their street looks like, what annoyed them at work. Scripted personas tend to answer in generalities or bounce the question back to you.
• Forgets what you said. Details you shared a few days ago are gone, or are remembered slightly wrong.
• Moves in stages. Compliments, then a shared "connection", then intimacy, then a request, arriving on a suspiciously neat timetable.
• Mirrors you too well. Likes exactly what you like, agrees with everything, has no opinions of their own.

When you ask for proof

• Refuses or keeps postponing a live video call. Camera broken, bad network, "not ready yet". This was exactly the pattern in the report.
• A call that feels off. Very short, no sound, face half hidden, or a clip that does not respond to what you just said. Ask them to do something specific and ordinary on camera, like wave with the left hand or hold up today's newspaper.
• Photos that are too perfect or that turn up elsewhere under another name.

When it turns toward money

• Scripted urgency. A crisis, a deadline, a reason you must act today.
• Pushing you to another app. "Let's talk on WhatsApp" or Telegram, very early. See why scammers rush you to WhatsApp.
• Crypto or an "investment opportunity". A new friend who happens to know a trading platform is one of the oldest scripts there is, and AI makes it cheaper to run at scale.
• Pay-to-keep-talking. If a conversation only exists while you keep buying credits, ask yourself who benefits from it continuing.

If a match asks for money, gift cards or crypto, you can stop reading signs. That alone is enough to end it. If you have already paid or are being threatened, go straight to what to do if you are being blackmailed and call 1930.

Why a live selfie reviewed by a person matters

The operation in the report worked because nothing stood between a generated identity and a user's feed. Profile photos can be generated, and uploads can be stolen. What is much harder to fake at scale is a live camera capture that a human being looks at before the profile goes public.

That is not a magic fix. A selfie check confirms that a real person was in front of a camera when the account was made. It does not confirm who is typing later, whether they are married or single, or whether they are kind. But it removes the cheapest trick in the report, which was thousands of personas that never had a face behind them.

AI companion vs AI persona: the difference is honesty

Not every AI on a dating platform is a scam. The line is disclosure.

	Disclosed AI companion	Undisclosed AI persona
Do you know it is AI?	Yes, it is labelled	No, it pretends to be a person
Counted as a member?	Should not be	Used to inflate the numbers
Your consent	You chose to talk to an AI	You were deceived

Talking to a labelled AI is a choice an adult can make, and whether it counts as cheating is a separate question we look at in Is an AI relationship cheating?. Being led to believe a program is a person you might one day meet is deception, whether or not money changes hands.

How Parda handles this

On Parda, every profile passes a live camera selfie check reviewed by a person before any member can see it, and there is no upload option. An optional government photo ID check adds an "ID verified" badge you can filter by. Neither check verifies marital status, criminal record or behaviour, and we say so in how verification works. Parda does have AI chat companions; they are clearly labelled AI and are never counted as members.

Whatever app you use, the rule is the same: ask for a live video call before you invest feelings, and walk away the moment money enters the chat.