Scammers Are Using AI to Fake Relationships

Scammers are leveraging AI to automate romance fraud, creating more believable fake profiles and conversations to manipulate and exploit their targets at scale.

Criminals are increasingly weaponizing artificial intelligence to automate and enhance romance scams, creating a new wave of what Malwarebytes Labs calls "artificial intimacy." These scams have long relied on social engineering to build false relationships and manipulate victims into sending money. However, the introduction of generative AI tools has made these fraudulent operations more sophisticated, scalable, and difficult for the average person to detect. Scammers are no longer limited by their own writing skills or the need to find stolen photos for their fake profiles. Instead, they can now generate entire personas, complete with unique backstories and convincing conversational styles, all powered by AI. This allows a single operator to manage multiple fraudulent relationships simultaneously, each tailored to the specific target, dramatically increasing the potential for financial and emotional damage. The core deception remains the same, but the tools used to execute it have become exponentially more powerful, posing a significant threat on dating apps, social media, and private messaging platforms.

At a technical level, these AI-powered scams leverage a combination of large language models (LLMs) and image generation tools. Scammers use LLMs like GPT-4 to craft initial outreach messages, maintain ongoing conversations, and write compelling, emotionally charged scripts designed to build trust and urgency. These models can adapt their language, tone, and personality in real-time based on the victim's responses, creating a highly personalized and believable interaction that bypasses the classic red flags of poor grammar or unnatural phrasing common in older scams. For the visual element, AI image generators like Midjourney or Stable Diffusion are used to create an endless supply of attractive and unique profile pictures. Because these images are entirely synthetic and do not exist anywhere else online, they cannot be identified through a reverse image search, a common method for verifying a person's identity. This combination of fluent, adaptive text and original, untraceable images creates a fraudulent identity that is nearly indistinguishable from a real person, dismantling key defenses that users have been taught to look for.

The rise of AI in romance scams is part of a much broader trend in which generative AI is becoming the tool of choice for all forms of social engineering. The same techniques used to fake a romantic interest are also being applied to more sophisticated business email compromise (BEC) attacks, personalized phishing campaigns, and the spread of disinformation. In each case, AI lowers the barrier to entry for attackers, allowing less-skilled individuals to execute highly convincing campaigns that were once the domain of well-resourced groups. This democratization of advanced social engineering tools presents a fundamental challenge for security professionals and platform moderators. The digital trust model, already fragile, is further eroded when users can no longer be certain if they are interacting with a human or a malicious AI. As these tools become more integrated into criminal workflows, the volume and sophistication of deception-based attacks across all sectors are expected to grow significantly.

For security teams and the companies that run online platforms, this trend necessitates a strategic shift in fraud detection. Relying on static indicators like IP blacklists, known scammer profiles, or simple keyword filtering is no longer sufficient. The defense must now evolve to focus on behavioral analysis and anomaly detection. This means developing systems that can analyze patterns in communication, such as response timing, linguistic complexity over time, and network-level activity, to identify non-human behavior. An arms race is now underway, with security companies developing their own AI models to detect the patterns left by malicious AI. Looking ahead, the next frontier for these scams will likely involve AI-generated voice and video. As deepfake technology becomes more accessible, scammers will be able to escalate their deception from text-based chat to seemingly real phone calls and video chats, making the illusion of intimacy even more powerful and the scams even harder to expose.

AI-generated text and images make it harder for security systems to flag fraudulent accounts based on traditional red flags like poor grammar. This requires developers and security teams to build new detection models that can identify sophisticated, AI-driven social engineering tactics.

Platforms like dating apps and social media sites face significant reputational damage and user trust erosion from AI-powered scams. The inability to effectively police this activity can lead to user churn, increased operational costs for moderation, and potential legal liability.