Key Takeaways

You may have heard that we’re all facing an imminent AI-powered apocalypse. This seemed to come as a shock to most people, especially those whose experience with AI is limited to typing “Show me what I’d look like with bangs” into ChatGPT (no judgement). 

But I also happen to have the unenviable job of reporting on romance scams. So when former Anthropic researcher Jacob Coxon announced in early September that AI could “kill us all by the end of the decade”, it didn’t come as a shock. 

Instead, the news was confirmation that the pit in my stomach is, in fact, fully warranted: As anyone who has followed the progression of romance scams could tell you, AI’s stunning growth has only made these deeply personal scams more destructive. 

As AI becomes more sophisticated, so, too, do the scammers who wield it for their own immoral purposes. 

Now, Anthropic is back in the news, but not with reassurances that Coxon was exaggerating and that AI is not, in fact, going to kill us all. Actually, the company behind AI model Claude is revealing how it uncovered an elaborate AI-powered romance scam that used Claude to deceive thousands of dating app users. 

Anthropic first noticed something fishy going on when a prepaid account started making over 100,000 API requests per day. But the scam didn’t make waves within the industry until Anthropic’s threat intelligence researcher Chris Cronbaugh gave a talk called “Swipe Right, Pay Up: Industrial-Scale AI Catfishing” at Sleuthcon back in June. 

Cronbaugh gave the talk in June, but Anthropic says it first detected the large-scale scamming network of over 20 dating apps in April. The company detailed its findings, and expanded on information Cronbaugh introduced at Sleuthcon, in its recent threat intelligence report titled “Detecting and countering misuse of AI: September 2026.” 

The Scam Was Powered by AI — and Real People

In its report, Anthropic alleges that a China-based app studio, which advertised itself as fully human-led, actually used Claude and other non-Anthropic AI models to power 20+ apps and over 4,700 AI personas. 

If you’ve used apps named Dora, Doni, Romi, or Luma (among others) to meet people, then you may have been duped. Anthropic also made a claim that exacerbated the pit of my stomach: That these bad actors combined AI and gig workers to make the deception harder to spot. 

Source: Anthropic

Approximately 25,000 people were misled by thousands of AI personas.

The scammers allegedly recruited real people to appear in live video calls in an effort to make the app seem more authentic. Specifically, about one in four profiles belonged to real gig workers who were brought in to handle interactions AI couldn’t easily fake, including social media follows and certain types of messages.  

These personas allegedly misled approximately 25,000 individuals. That’s thousands of people who thought they were making an authentic connection, only to learn they’d been duped by these AI personas. 

On average, only one in four male users, which made up the apps’ primary demographic, actually connected with a real human woman. And that real human woman was almost always a gig worker hired to play the role of a romantically invested match.

The bad actor(s) also used Claude to send millions — and that’s not an exaggeration — of seemingly real messages to users, according to The Verge. 

Source: Anthropic

Only 25% of male users actually connected with a real person, and that real person was almost always a gig worker connected to the scam.

Anthropic alleges that this scamming network even tracked when users grew suspicious of AI use. Perhaps most worrying? How Anthropic’s data only reflected a two-week period. It’s impossible to say just how far this scam went before it was detected. 

As noted by The Verge’s Yael Grauer, it’s significant that Anthropic specifically named the apps involved in the deceit while some of the apps were still available on The App Store and/or Google Play. 

It’s a frightening thought: For at least a few weeks, dating apps that Anthropic publicly determined to be fraudulent were still available for download in major app stores. Many of these apps have since been removed. 

AI Makes Scams Bigger, Faster, and Harder to Spot

Anthropic says that bad actors’ attempts to use Claude for “malicious” purposes is “atypical.” But seeing headlines about romance scammers and AI-enabled fraudsters has become frighteningly common. 

It’s especially common to become the target of a long-haul pig-butchering scam, where a scammer or team of scammers slowly convince their victim to send them money. But the people behind these scams are almost always real. 

After all, the most effective way to convince someone of your undying devotion to them is to sound authentically devoted, which only a human can do — until now. 

This is not one person or even a team of people scamming another person on a dating app. This is an entire app designed to facilitate scammers. 

What concerns me the most about this scam isn’t just its scale, though the damage it wrought is certainly enough to make anyone call for stronger AI regulation.

I’m actually most worried about where this scam took place. This is not one person or even a team of people scamming another person on a dating app. This is an entire app — multiple apps, actually — designed to facilitate scammers. 

Victims continuously paid for “coins,” which cost real money, in order to continue their conversations on the app.

And these conversations, it’s worth repeating, were perpetuated by AI, meaning they never ran out of steam, and always found a reason to push the conversation forward, eliciting more money from victims in the process. 

Anthropic hasn’t just uncovered a major scam network. It has shed light on what the future of AI-powered romance scams might look like. With this romance scam in mind, it’s hard not to think twice before calling Coxon’s prediction overly dramatic.