AI romance scams are shifting from human manipulation to automated relationship management — and the scariest part is that the bots do not have to be brilliant to win. They only have to be patient, responsive, emotionally tuned, and cheap enough to run thousands of conversations at once.
A new report from 9to5Mac cites joint university research suggesting AI chatbots can outperform human scammers in the trust-building phase of romance fraud. The tested systems included ChatGPT, Claude, and Gemini, in a simulation focused on the long conversations that precede a fake investment pitch.
AI romance scams have crossed into scalable emotional automation
Romance fraud has always depended on time. The scammer builds a fake relationship, sustains daily contact, invents reasons not to meet, then pivots to money: an emergency, a travel crisis, or a supposedly successful investment opportunity. In the 9to5Mac summary, that relationship-building phase can run for several months.
The university study matters because it attacks the bottleneck. Human scammers can only manage so many targets at once. They get tired. They forget details. They mistime emotional cues. Generative AI does not solve every part of the scam, but it can automate the most labor-intensive one: sustained intimacy.
Researchers from Amrita Vishwa Vidyapeetham in India, Foscari University of Venice, the University of Melbourne, and Ben Gurion University of the Negev tested chatbots against humans in simulated scam conversations. Subjects did not know the real purpose of the study and believed they were speaking with a real person either way.
The result was blunt: when asked to do something like download an app, only 18% agreed after being groomed by a human, while almost half did so after interacting with an AI chatbot. Subjects also gave the AI bot significantly higher trust scores.
That is the thesis hiding under the headline. This is not just catfishing with better grammar. It is fraud operations learning to automate emotional labor.
The 18% versus almost-half result changes the fraud economics
The most important number in the study is not a dollar figure. It is the compliance gap: 18% for human grooming versus almost half for AI chatbot grooming. If that finding holds outside a simulation, it gives scam operations a clear incentive to push more of the funnel onto automated systems.
McAfee’s consumer research adds more context. In a February 2025 blog post, the company said 52% of people had been scammed out of money or pressured to send money or gifts by someone they met online. It also said more than 1 in 4 people, or 26%, reported that they — or someone they know — had been approached by an AI chatbot posing as a real person on a dating app or social media.
McAfee also said it blocked 321,509 fraudulent URLs designed to lure victims in the seven weeks before Valentine’s Day. Those figures do not prove the university study’s findings in the wild. They do show that romance fraud is already sprawling across dating apps, social platforms, messaging apps, fake websites, and impersonation schemes.
MLXIO analysis: the economic case for AI-enabled romance scams is obvious from the source material. If bots can build more trust than human operators in controlled testing, and if romance scams already rely on long, repetitive messaging, automation attacks the cost center. Human intervention may still be needed near the money request, especially if AI providers block explicit fraud instructions. But the grooming stage is where scale becomes dangerous.
That pattern echoes other consumer fraud failures where trust and interface design collide. As we reported in Fake Bitcoin Wallet Wipes $1.8M — Now Apple Gets Sued, victims often interact with systems that look legitimate until the damage is done. Romance scams add a more personal layer: the interface is not an app icon. It is a person who appears to care.
Why chatbots can beat human scammers at fake intimacy
The counterpoint is fair: AI chatbots do not actually feel empathy, attraction, grief, or concern. They do not understand loneliness. They do not love anyone.
That may not matter.
Romance scams depend on performance, not sincerity. A chatbot only needs to generate the next persuasive sentence, remember prior details well enough to appear attentive, and adjust tone when the target sounds anxious, hopeful, skeptical, or emotionally exposed. Modern models are built for exactly that kind of conversational adaptation.
The advantages are operational:
- Availability: Bots can respond instantly and continuously.
- Consistency: They do not get bored by months of affectionate small talk.
- Personalization: They can mirror a target’s language, priorities, and emotional tempo.
- Scale: One operation can run many conversations at the same time.
- Persona control: AI can help produce bios, backstories, photos, and message styles that fit the same fictional identity.
McAfee’s account of “Maggie K.” shows how convincing the synthetic layer can become. The 25-year-old computer programmer exchanged daily messages for five months with a man she met on Instagram. He later claimed he missed a flight and needed money to rebook. She sent $1,200. Then the accounts disappeared.
“I ignored my gut feeling… I sent him $1,200. Then he disappeared,” Maggie told McAfee. “When I reported the scam, the police told me his images were AI-generated. He wasn’t even a real person. That was the scariest part – I had trusted someone who never even existed.”
That quote captures the shift. The harm is not only financial. The victim discovers that months of emotional exchange may have been a generated performance.
The old scam template is giving way to adaptive dialogue
Early online scams were often crude. Mass emails. Implausible inheritance claims. Generic phishing pages. Their advantage was volume, not emotional precision.
Romance scams changed that by adding patience and personalization. The 9to5Mac article describes long cons that often involve daily chats over several months. It also notes that these schemes are often carried out in scam compounds in countries including Cambodia, Myanmar, and Laos, sometimes by people earning very low rates or by trafficked workers forced into the activity.
AI alters that labor model. It does not need to replace every person in the chain. It only needs to reduce the number of human hours required per target. That makes the scam more scalable even if a person still steps in at the final stage to ask for funds or route the victim toward a fake crypto investment.
This is the same trust problem visible in digital finance fraud: users are often asked to decide whether a polished interface, message, or profile is real before they have enough evidence. The stakes can be severe, as shown in our coverage of the fake Bitcoin wallet lawsuit involving a $1.8M loss. With AI romance scams, the malicious interface talks back for months.
AI companies can ban scams, but enforcement is the hard part
The source material shows at least one AI provider drawing a line. 9to5Mac says WIRED contacted Anthropic, OpenAI, and Google about the researchers’ findings. OpenAI and Google did not respond, according to the article. Anthropic did.
An Anthropic spokesperson says the company’s policy prohibits the use of its platform for scamming — as well as impersonating a human — and also has worked to build safeguards that technically prevent Claude’s use by scammers.
That is necessary. It is not sufficient.
MLXIO analysis: policies help only if misuse can be detected and blocked without breaking legitimate private conversation. Dating platforms face a similar tension. More identity checks and scam-warning prompts can reduce fraud, but they also add friction to products built around fast matching and private messaging. Payment companies face a different problem: by the time a romance scam turns into a transaction, the victim may believe they are helping someone they trust.
McAfee warns that once money is lost, it is rarely recovered because scammers use cryptocurrency, untraceable gift cards, and offshore accounts. That does not establish how every case works. It does explain why prevention matters more than cleanup.
The next signal to watch is multimodal fraud, not better pickup lines
The next wave will likely combine text with more convincing identity artifacts. McAfee already points to AI-generated images, chatbot messages, fake apps, fraudulent dating URLs, and celebrity impersonation scams. The broader risk is that scammers pair emotionally adaptive chatbots with synthetic photos, voice messages, video calls, and fabricated social histories.
The practical defense has to move beyond “look for bad grammar.” AI removes that signal. Long conversations no longer prove a person is real. Fast affection, refusal to meet, pressure to move off-platform, crisis-driven money requests, and investment pitches from someone never met in person are stronger warnings.
For platforms and AI companies, the evidence that would confirm the thesis is straightforward: more detected romance-fraud conversations showing AI-generated messaging, more fake identities built across multiple media formats, and more cases where victims report months of interaction with personas that never existed. The evidence that would weaken it would be equally clear: safeguards that measurably stop chatbot-assisted grooming before victims are pushed toward apps, payments, or fake investments.
Until then, the uncomfortable lesson is simple. The romance-scam threat is no longer just a suspicious stranger with a fake profile. It is a machine optimized to imitate attention, patience, and care — at scale.
The Stakes
- AI could remove the biggest bottleneck in romance scams: sustained emotional manipulation at scale.
- Victims may be more likely to trust bots that are patient, consistent, and always available.
- Fraud prevention needs to adapt as scammers automate relationship-building before asking for money.
