一名操作者經營數千段虛假戀情：AI 欺詐的新經濟學

When technology companies discuss security, they usually focus on external threats like data breaches, vulnerable infrastructure, or stolen credentials. But for businesses built around human interaction, one of the most serious threats can exist inside the product itself, in the accounts users believe belong to real people. This is particularly relevant to online dating, where people share personal information, invest time in conversations, and develop emotional connections with strangers. The consequences of fraud extend beyond emotional connections with strangers. They affect users’ willingness to trust the platform and the people they meet through it.

According to research published by Sumsub in 2026, 84% of surveyed UK dating app users believed AI-generated content had made it harder to trust potential matches or date successfully, compared with 64% in 2025. Another 61% had experienced deception through fake profiles themselves or know someone who had. These findings reflect consumer perceptions rather than the actual prevalence of fraudulent accounts. Nevertheless, they reveal a problem for an industry whose success depends on people being willing to interact with strangers. AI is making fake engagement cheaper and that creates a fundamental challenge for businesses built on genuine human interaction.

How AI Changed the Economics of Fraud

Romance scams are not new. Fraudsters have used stolen photographs, fabricated identities, and scripted conversations for decades. What has changed is the cost and complexity of operating these schemes at scale. Previously, maintaining thousands of convincing profiles required significant human resources and scammers needed to manage conversations, remembering personal details and maintaining consistent identities and responding to users over extended periods. Generative AI has reduced much of that operational burden.

In September 2026, Anthropic published an investigation into a network of more than 20 deceptive dating applications operated by a China-based studio. During a two-week period, researchers identified over 4 700 AI persons’ interactions with at least 25 000 people and generated approximately 2.36 million messages. The operation combined AI-generated conversations with human workers who handled interactions requiring a real person, including live video calls. The applications concealed the automated nature of the service while presenting users with what appeared to be genuine dating opportunities. There were applications designed around deception rather than fraudulent profiles operating inside established dating platforms. However, the case illustrates how dramatically AI can reduce the resources required to manufacture convincing human interactions.

For legitimate dating platforms, similar technologies create a growing challenge. Generated photographs, automated messaging, stolen identities, and tools designed to bypass verification systems can be used to create and maintain fraudulent accounts. The problem is no longer limited to identifying suspicious profiles. Platforms must respond to increasingly sophisticated attacks while the cost of producing them continues to fall. And it changes the economics of operating a dating business. Trust is both a product and business problem.

Fraud carries business costs beyond moderation. Fake accounts generate registrations, likes, and conversations that look legitimate in product analytics, which can distort the signals behind recommendation algorithms and A/B tests. There is a commercial side as well – users who keep up meeting fake profiles may stop starting conversations, stay less active or pay less. Though the size of that effect varies by platform and must be measured rather than assumed. In dating, anti-fraud is therefore part of product economics not just a security function and as AI makes synthetic activity cheaper to produce, telling real engagement from manufactured engagement becomes a core product decision.

The Management Dilemma: Security Versus Growth

Stronger verification seems like an obvious response to increasingly sophisticated fraud. In practice, the decision is more complicated, every additional step in onboarding introduces friction. Requiring identity documents, biometric verification, or repeated checks may prevent certain forms of abuse, but it can also discourage legitimate users from completing registration. That directly affects acquisition economics. If the same marketing budget produces fewer registered and activated users, acquisition costs per completed registration increase.

For dating apps, the consequences can extend beyond the onboarding funnel. Unlike many other consumer products, dating platforms depend on local network density, where users need enough relevant active potential partners within a particular geography for the matching experience to work. A verification process that unnecessarily limits legitimate registrations can therefore slow the development of a local network, reducing matching opportunities and affecting retention. At the same time, allowing fraudulent accounts to remain active introduces other costs: infrastructure consumption, moderation workload, distorted recommendations, and declining user confidence. Excessive security can make a product difficult to use. Insufficient security can make it impossible to trust.

Fraud prevention should be managed as a product and business decision, not an isolated technical function. A risk-based approach combines basic protections at registration with behavioural analysis, device signals, content monitoring and stronger verification when suspicious activity appears. Even so, passing a verification check doesn’t make a user trustworthy, because proving presence does not prove intent. Effectiveness should be judged across several dimensions together, including detection performance, false positives, enforcement speed, verification costs, conversion, and retention. Better detection that wrongly restricts many genuine users is not a better business outcome. Privacy adds another layer, since analyzing communications or collecting extra identity signals requires transparency, data minimization, and strict access controls. The question is not how many fraudulent accounts a platform can block, but how well it reduces risk while preserving a good experience for legitimate users.

Beyond Verification: The Role of Community

As AI-generated conversations grow more convincing, a product question emerges. Should dating platforms focus only on digital interactions, or also create more chances for them to develop in the real world? At Fledge, we are exploring a model that pairs compatibility-based matching with offline events and community experiences, so that the step from algorithmic discovery to meeting in person feels natural.An event does not prove anyone is trustworthy or replace verification, but it adds social context that profiles and chats cannot reproduce. We see it as a promising direction, not a standalone answer to AI fraud.

AI should help people find compatible matches, support moderation, and detect suspicious behaviour, not manufacture the appearance of relationships. Success should therefore be measured beyond active users, matches, and conversion, adding interaction quality, authenticity, and the outcomes users actually seek. A platform with heavy engagement but few genuine connections risks optimizing for the wrong result.

Trust as a Competitive Advantage

AI will keep strengthening both fraudsters and the platforms defending against them and no single technical solution is likely. Protection will take continuous adaptation, multiple detection layers, responsible and careful evaluation of business trade-offs. But the challenge goes beyond cybersecurity. Trust shapes product quality, network growth, user behavior, and long-term revenue, and as fake engagement gets cheaper, authentic interactions become more valuable. In dating, trust is not a safety feature and, again, trust is the product. The next generation of platforms may win less by generating more interactions and more by building an environment where users can trust those interactions are real.