AI Scammers Can Earn Your Trust Before They Ever Ask for Money

By "Tracey, the Safety Lady" Hawkins, Safety and Gen AI Cybersecurity Expert for End Users

The next message that starts a costly scam probably won't have a misspelled word, a suspicious attachment or an urgent demand for cash.

It might show up as a friendly text from someone who claims to have the wrong number. It could land through a dating app, a social media account, a professional group or an online game.

The sender keeps the conversation going, asks good questions and remembers details you mentioned days earlier. Nothing alarming happens at first, and that's exactly the problem.

A July 30 WIRED report examined new research on how well artificial intelligence can handle the relationship-building stage of an online scam, the part that used to take a human weeks to pull off. In a weeklong experiment, 22 participants exchanged messages with a human romance-fraud expert and with an AI agent built on an older Claude model paired with GPT-4o, all while believing both were real people.

By the end of the week, 46% of participants agreed to download an app the AI recommended. Only 18% agreed to download a game the human suggested. Participants rated the AI as more trustworthy and sent it 80% of their messages. Only one person figured out during the experiment that the more engaging texter was a machine.

It was a small study, and downloading an app isn't the same as losing money, so those limitations matter. But the results still point to something worth taking seriously: an AI-powered scammer may be able to earn someone's confidence before money, passwords or personal information ever come up. The scammer doesn't have to rush anymore A lot of people assume scams are easy to spot because criminals are pushy. That assumption is exactly the blind spot AI is learning to exploit.

Relationship-based scams often unfold slowly. A criminal might spend weeks building a friendship or romance before ever mentioning an investment.

By the time the pitch arrives, the victim isn't weighing a financial offer from a stranger. It's coming from someone who feels like part of their daily life.

The researchers behind this study also interviewed 145 former scam workers, including trafficking survivors who had been forced to labor in scam compounds. Those interviews described a three-stage playbook: hook the target, spend a long stretch building trust, then close with a fraudulent investment ask.

AI is what makes the middle stage scalable. A chatbot can carry on conversations all day, track personal details and generate responses that sound supportive, which means one criminal can now run far more targets at once because software is doing most of the talking. That software doesn't feel friendship or concern. It produces language convincing enough to fake it, and that distinction is nearly impossible to catch from words on a screen.

This happens to smart, careful people

Victims of relationship and investment scams get judged after the fact. Friends and family wonder how anyone could believe the story or send the money. That reaction misunderstands how these crimes actually work.

Criminals don't open with the unbelievable part. Early messages are designed to feel ordinary. Personal details get collected gradually, and shared interests and emotional connection get built before any financial request shows up. AI just makes that personalization sharper.

Someone who mentions feeling lonely gets more companionship. Someone worried about retirement hears about an investment opportunity. Someone job-hunting gets approached with a role.

The conversation bends around whatever the target reveals.

These scammers aren't trying to overpower good judgment. They're patiently working to lower it, which means fraud prevention has to account for the grooming period, not just the moment money changes hands.

Waiting until someone asks for money is often too late, because by then the victim has already spent weeks accepting the sender's identity as real. Watch the request, not the writing quality

Old scam advice focused on how a message looked: bad grammar, odd spelling, awkward phrasing. Those signs can still matter, but polished writing no longer proves a sender is legitimate.

Generative AI produces clean, natural-sounding text in seconds and can keep a conversation going indefinitely without the slip-ups people are trained to look for.

Pay closer attention when an online contact asks you to:

Download an unfamiliar app or piece of software Move the conversation off the original platform Invest in cryptocurrency or use a trading site they recommend Send money, gift cards or account information Share a password, login code or ID photo Keep the relationship or opportunity secret Avoid telling family, friends or a financial professional Give someone remote access to your phone or computer

None of those requests get safer just because the relationship has lasted a few weeks.

How to protect yourself

Start with one rule: trust earned online is not identity verification. Look up companies, investment platforms and apps on your own. Don't rely on a website, phone number or contact info the other person hands you. Talk to someone you trust before you move money or install software tied to an online relationship, because that person hasn't spent weeks absorbing the attention that may be clouding your judgment. And don't assume a video call settles anything. Stolen footage, face-swapping tools and deepfakes can fake visual contact without faking the person behind it. Refuse any request for passwords, verification codes or remote device access.

Real financial institutions and government agencies don't need those to help you.

If something feels off, take screenshots, preserve the messages, call your financial institution right away, report the account to the platform and file a report with law enforcement. Shame protects the criminal. Reporting protects the next person.

This isn't just a dating problem

AI-assisted relationship scams aren't limited to romance. A fake client can spend weeks building rapport with a real estate agent before producing fraudulent ownership documents. A supposed vendor can befriend an employee before requesting a change in payment instructions.

The same playbook works on a fake executive, donor, customer or recruiter.

Verification can't depend on whether someone sounds credible. Financial changes, new account details and sensitive requests need confirmation through a known phone number or a separate trusted channel, every time. And families and workplaces need to talk about this without ridicule, because someone is far more likely to ask for help early when they know they won't be mocked for it.

Why this matters to your bottom line

This is exactly the gap I built my Gen AI Cybersecurity Certificate programs to close, for real estate professionals, attorneys, property managers and the businesses that work with them every day. AI-powered social engineering doesn't just put individuals at risk.

It exposes companies to fraud losses, lawsuits, and reputational damage that can take years to rebuild.

One successful scam against an employee or a client can cost far more than the price of training a team to recognize the pattern before it costs them anything.

I teach people and organizations how to use AI without getting scammed, sued or shut down, and that training pays for itself the first time it stops a six-figure wire transfer or a lawsuit tied to a preventable breach.

If you want your team, your association or your clients equipped to spot this before it costs them, that's the conversation worth having now, not after the first loss.

A warm conversation isn't proof. A familiar name isn't proof. Before you download the app, share the information or send the money, step outside the relationship and verify independently.

About the author: "Tracey, the Safety Lady" Hawkins is the gen AI cybersecurity expert for end users who protects people, profits and reputations. She teaches people and organizations how to use AI without getting scammed, sued or shut down.