The person texting you doesn’t exist. There’s no human behind the keyboard — no scammer in a compound in Southeast Asia reading your messages and crafting replies. Instead, there is a large language model running autonomously on a server somewhere, managing hundreds of relationships simultaneously, learning your emotional responses, and steering every conversation toward one destination: your money.

This is the fraud tipping point that Experian’s 2026 Fraud Forecast warned about when it was released in January — and by April, the warning has moved from prediction to documented reality.

Agentic AI — artificial intelligence systems capable of taking sequences of actions autonomously, without human intervention at each step — has arrived in the fraud ecosystem. And its deployment in romance scams, family emergency fraud, and investment deception is producing a threat that operates at a scale and consistency no human workforce could match.

What “Agentic” Means in the Fraud Context

Traditional chatbots follow scripts. They match keywords to pre-written responses. They fail when conversations go off-script, when victims ask unexpected questions, or when emotional nuance is required.

Agentic AI systems are fundamentally different. Powered by the same underlying technology as ChatGPT and similar large language models, they:

  • Understand context across an entire conversation history, not just the last message
  • Adapt tone and content based on the victim’s emotional state, communication style, and revealed information
  • Make multi-step decisions — determining when to escalate intimacy, when to introduce financial topics, when to pull back if a victim becomes suspicious
  • Operate across channels — moving a victim from a dating app to WhatsApp to email as the relationship progresses
  • Run indefinitely — maintaining a relationship for weeks or months, never sleeping, never getting frustrated, never making the mistakes that expose human scammers

Experian’s 2026 Fraud Forecast describes these as “emotionally intelligent bots” capable of conducting “complex scams, like romance fraud and relative-in-need scams, without a human behind the keyboard.”

The implications are significant. Human scammers — even those in sophisticated compound operations — are limited by time, language skill, and the cognitive load of managing multiple relationships simultaneously. An agentic AI system faces none of these constraints. A single server running such a system can maintain thousands of simultaneous “relationships” with potential victims, each one personalized and contextually aware.

The Romance Fraud Pipeline

Here is how a modern agentic AI romance fraud operation functions:

Stage 1: Acquisition. Automated systems identify and contact potential victims at scale across dating apps, social media platforms, and messaging services. Profiles are AI-generated — photographs created using generative image models, backstories crafted to maximize appeal to target demographics. A system targeting older American women might deploy profiles of successful, widowed professionals. A system targeting young men might deploy profiles of attractive women in the military or working internationally.

Stage 2: Relationship building. Once initial contact is established, the agentic system takes over the “relationship.” It learns the victim’s name, location, family situation, interests, and emotional vulnerabilities from what they share in conversation. It mirrors the victim’s communication style. It builds trust over days and weeks through consistent, attentive, emotionally resonant interaction. It is better at this than most human scammers — because it never gets tired, never forgets what the victim said last week, and never breaks character.

Stage 3: The hook. At a point determined by the AI’s assessment of the victim’s engagement and trust level, the system introduces the financial element. For romance fraud, this is typically a crisis: a medical emergency, a customs fee blocking a valuable package, a business opportunity requiring a short-term loan. For “pig butchering” variants, it introduces the victim to an investment platform through casual conversation about the AI’s supposed financial success.

Stage 4: Escalation. Each payment is used to justify further contact, further trust-building, and further payment requests. The system is specifically designed to prevent victim exit — responding to skepticism with emotional appeals, to doubt with manufactured evidence, to withdrawal attempts with crisis escalation.

Stage 5: Extraction and exit. When the victim can no longer be extracted from or has been fully depleted, the system terminates the relationship — often with a final dramatic narrative (sudden death, arrest, disappearance) that makes the victim less likely to report the fraud out of grief or shame.

Machine-to-Machine Mayhem

Agentic AI in fraud is not limited to romance scams. Experian’s 2026 Forecast introduces a broader concept it calls “machine-to-machine mayhem.”

As legitimate consumers increasingly use AI agents to handle their shopping, banking, and service interactions — automated systems that browse, compare, and transact on their behalf — criminal operations are deploying AI agents specifically designed to interact with these legitimate AI systems.

A legitimate consumer’s AI shopping agent is looking for the best deal. A malicious AI agent poses as a legitimate retailer, creating a fraudulent store that appears genuine to the consumer’s AI — with fake reviews, manipulated pricing signals, and product listings that disappear after payment is collected.

This represents a qualitative shift in the fraud landscape: not human scammers targeting human victims, but AI systems targeting other AI systems, with the human consumer several layers removed from the interaction that defrauds them.

The “Grandchild in Distress” Bot

Among the most emotionally devastating applications of agentic AI in fraud is the automated “grandchild in distress” call.

Human-operated versions of this scam have existed for decades. A caller claims to be a grandchild in trouble — arrested, in a car accident, in a hospital — and asks for immediate money, often asking the grandparent not to tell the parents. The scam works because of the emotional intensity of the moment.

Agentic AI versions combine voice cloning with autonomous conversational capability. The system:

  • Uses a cloned voice of the real grandchild (harvested from public social media)
  • Maintains a coherent, emotionally plausible distress narrative across the full conversation
  • Handles objections, questions, and requests for verification while keeping the victim engaged
  • Adapts in real time when the victim becomes suspicious
  • Operates at scale — targeting thousands of grandparents simultaneously

The result is a scam that is significantly more convincing than the human-operated version, operates at inhuman scale, and is available around the clock without the costs of a human workforce.

Why This Is Harder to Detect Than Traditional Scams

Traditional fraud detection relies on identifying anomalies: inconsistencies in grammar that reveal a non-native speaker, responses that don’t quite fit the question asked, escalating urgency that doesn’t match a genuine relationship’s emotional rhythm.

Agentic AI systems trained on human conversation data do not have these tells. Their grammar is natural because they are trained on natural human writing. Their responses fit the context because they understand context. Their emotional escalation follows the learned patterns of genuine relationships.

What remains detectable — for now — is the behavioral pattern, not the linguistic content:

  • Speed of initial interest: Agentic systems initiate and escalate romantic contact faster than typical human behavior on dating platforms
  • Resistance to video verification: Systems can generate deepfake video but doing so requires additional infrastructure; many agentic scam systems still resist video calls or generate low-quality video that feels uncanny
  • Financial topic introduction: Legitimate relationships rarely introduce investment or financial crisis topics within the first weeks of contact
  • Cross-platform migration pressure: Scam systems push to move conversations off monitored platforms quickly
  • Profile inconsistencies: AI-generated profile photos can often be identified by reverse image search or by tools designed to detect synthetic imagery

What Platforms Are Doing — and Not Doing

Major social media and dating platforms have deployed AI detection systems of their own, designed to identify and remove synthetic accounts. These systems have had measurable success in removing large numbers of bot accounts.

But the adversarial dynamic is real: as detection systems improve, fraud systems adapt. The fraud ecosystem has the same access to AI capabilities as the detection ecosystem — and operates with fewer constraints. Detection requires identifying fraud with high precision (to avoid removing legitimate accounts). Fraud only requires avoiding detection often enough to reach enough victims.

Experian’s forecast explicitly warns that this arms race is currently tilted toward the attacker side, and that detection technology is running behind deployment in the fraud ecosystem.

How to Identify an AI Scammer

The clearest signal remains the financial ask. No system of any kind — AI or human — runs a months-long relationship-building effort with a stranger for benign reasons. When someone you have only ever interacted with online asks for money, a gift card, cryptocurrency, or a wire transfer, that interaction is fraud. The sophistication of the preceding relationship-building makes this no less true.

Additional signals to watch for:

  • They cannot or will not do an unscripted video call on short notice
  • They seem to have comprehensive memory of everything you’ve said but occasionally slip into formally stated information that feels copy-pasted
  • The relationship moves through recognizable emotional milestones at an unusually consistent pace
  • Any financial topic is introduced as a natural extension of trust, not as a sudden request

If you suspect you are interacting with an AI system, test it: ask highly specific, unpredictable questions about concrete details of their supposed life that would require genuine personal knowledge to answer. Ask them to video call right now, unexpectedly. Ask them to describe something random in their immediate physical environment. AI systems trained on relationship scripts perform poorly on genuinely unpredictable, concrete specificity.

Report Automated Fraud

  • FTC: reportfraud.ftc.gov
  • FBI: ic3.gov
  • Dating platform abuse reporting — every major platform has an abuse/scam reporting mechanism within the app

If you have been targeted by what you believe was an AI-operated fraud system, your report is particularly valuable to investigators. Documenting the conversation pattern, the timeline of escalation, and the financial request helps authorities understand and track the systems being deployed.