TL;DR
Anthropic reported that a suspected dating-app operation used 4,700 personas generated with its Claude AI to contact 25,000 people and send more than 2 million messages over two weeks. The report says human workers supplemented the AI, while the operation made money through in-app tokens rather than direct requests for money. The full extent of losses and the identities of those responsible are not established in the supplied material.
Anthropic says it discovered an alleged dating-app operation that used 4,700 unique personas built with Claude to interact with 25,000 people, sending more than 2 million messages over two weeks. The report describes how generative AI and a limited number of human workers were combined to make dating conversations appear authentic, while monetizing them through tokens purchased in the apps.
According to Anthropic’s report, the people behind the operation worked with a China-based app studio to create a network of 20 dating apps and generate AI personas. The personas were used to draw users into extended conversations. Anthropic said a small number of gig workers were brought in for interactions that the software could not convincingly handle, including brief video calls and social-media follows.
The scheme differed from many romance scams described in the source material: rather than primarily persuading people to send money directly to an individual, the operators allegedly made money from tokens users bought to continue conversations. The report says users thought they were speaking with people, though the interactions involved both automated and human activity. It does not provide a total revenue figure or establish how much any individual user spent.
The report comes amid broader concern about online fraud. The original report cites nearly $21 billion in internet-crime losses in 2025 and nearly $1 billion attributed to romance scams, while noting that more than 64,000 Americans reportedly lost over $1 billion to romance scams in 2023. Those figures refer to different years and categories. McAfee, a cybersecurity company, separately found that 35% of Americans who use dating apps believed they had encountered fake photos or profiles; the supplied material does not specify the survey’s date or sample size.
How AI Changes Dating-App Scams
The reported operation matters because it shows how AI can scale the early stages of a deception: creating many profiles and sustaining large volumes of conversation. Human involvement in video calls and social-media activity, as Anthropic described it, also suggests that automated messages may be paired with real-person contact to make an account seem more credible.
The alleged token-based model points to a risk beyond familiar requests for emergency cash or bank transfers. Users may incur repeated charges while trying to maintain a relationship they believe is personal. That does not mean every intense or paid dating-app conversation is fraudulent, but the report underlines why profile authenticity and payment practices are relevant to user safety.
Dating services and payment companies have incentives to detect suspicious behavior, and the source says platforms including Bumble, Match, Tinder, OkCupid and Plenty of Fish use AI to identify fake profiles or harmful conduct. Visa also said its Scam Disruption team found a romance-scam network using dating sites and chatbots to steer people into recurring billing. These examples indicate industry responses, but do not establish how well any particular system prevents scams.
AI detection tools for dating apps
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A Shift From Direct Payment Requests
Romance scams commonly rely on a gradual relationship-building process before an account asks for money. The source describes a separate case involving six Nigerian men extradited to the United States, accused of using online romance to defraud more than 100 women of over $6 million. The allegations concern that case and should not be conflated with the AI operation Anthropic reported.
The source also says authorities recorded more than 64,000 American romance-scam victims in 2023, with losses exceeding $1 billion. These figures offer a baseline for the scale of reported fraud, but they do not show whether AI has caused an increase. The Anthropic account adds a different mechanism: automated personas, high-volume messaging and token purchases, with human workers filling selected gaps.
Some dating services and financial companies use AI to identify suspicious accounts and transactions, according to the source. PeopleFinders has also developed a dating-focused background-check product called Stud or Dud, which it says can review information such as employment, marital history and criminal records. Such services are tools, not proof that a person is safe or unsafe.
“The operation involved 4,700 unique personas interacting with 25,000 people and generating more than two million messages over a two-week period.”
— Anthropic, in its report on the alleged operation
As an affiliate, we earn on qualifying purchases.
What the Report Does Not Establish
The supplied account does not give the exact dates of the two-week operation, identify the app studio or operators by name, or say whether law enforcement has filed charges connected to this network. It also does not state how many users paid for tokens, the total amount collected, or how many people experienced financial harm.
Anthropic’s reported findings are attributed to the company; the source material does not include a response from the app studio, the alleged operators, or the dating services involved. The relationship between the 20-app network and named mainstream dating platforms is not specified. The available figures also do not show whether AI-driven schemes have increased romance-scam losses compared with earlier years.
cybersecurity tools for online scams
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Further Details and Platform Responses
Any law-enforcement action, company response or additional technical findings would clarify who operated the app network and whether users can seek refunds or other remedies. The source material provides no timetable for further disclosures or investigations, so those steps remain unconfirmed.
For now, dating platforms’ detection practices and payment protections are central to how the reported threat may be addressed. The source advises users to question inconsistencies and verify information rather than relying on a profile alone. Background-check tools may provide information, but the material does not establish that they can detect every fabricated persona or prevent fraud.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Anthropic report?
Anthropic said an alleged operation used 4,700 Claude-generated personas to interact with 25,000 people and send more than 2 million messages over two weeks.
How did the reported operation make money?
According to the report, users bought tokens to continue conversations. The supplied material does not give a total revenue figure or average spending per user.
Were the conversations entirely automated?
No. Anthropic’s account says AI generated personas and handled conversations, while a small number of human gig workers took part in activities such as brief video calls and social-media follows.
Is this operation linked to the six men extradited from Nigeria?
The source describes the extradition case separately. It does not establish a connection between those men and the AI-enabled app network Anthropic reported.
What remains unknown about the alleged scheme?
The source does not identify the operators or app studio, provide exact dates, state total token revenue, or describe any related charges. It also does not say how many users lost money.
Source: rss
