Here’s the thing about old-school phishing scams — you could usually spot them. Bad grammar, weird email addresses, some random message about a prize you never entered. Easy to ignore. But that’s not really how it works anymore.
AI chatbots have made it possible to craft scam messages that sound like they actually know you. And the raw material? Just your public social media. Your posts, your photos, who you follow, what you talk about — that’s enough to build something scary convincing.
Reuters worked with a Harvard researcher to test exactly this. They used popular AI chatbots to help plot a simulated phishing scam — writing the emails, figuring out the best timing, the whole thing — and then tested it on 108 volunteers. The AI didn’t just write a generic message. It helped make the attack feel personal and legitimate.
And it’s not just one chatbot. Researchers studying this threat tested four major commercial AI tools — ChatGPT (GPT-3.5 Turbo), GPT-4, Claude, and Bard — and found all of them could be used to generate phishing content. These are tools millions of people use every day for totally normal stuff. That’s what makes this tricky.
So what does a targeted AI phishing message actually look like? It might mention your school. Your job. A club you’re in. A trip you posted about. A person you actually know. It might come disguised as a message from your bank, your university, or even a friend — and it sounds right because the AI was fed real details about your life to work with.
The goal is always the same: get you to click a link, hand over a password, or send money. But when a message feels personal, your guard goes down. That’s exactly what makes this version more dangerous than the old spray-and-pray approach.
A few things worth keeping in mind:
Your public posts are literally the training data. Anything public — your name, school, city, interests, relationships — can be scraped and fed into an AI to personalize an attack against you. You don’t have to have done anything wrong. Just existing online with a public profile is enough.
The scam can come from anywhere. Email, DMs, texts. It might look like a message from a brand you follow, a school administrator, a job you applied to, or a financial service you actually use.
Speed is part of the problem too. AI doesn’t get tired. It can generate targeted messages at a scale and pace no human attacker ever could — meaning more people get hit, faster.
The best defense is actually pretty simple: treat any unexpected message asking you to click something, log in, or send anything as suspicious — even if it sounds like it knows you. Especially if it sounds like it knows you. Go directly to the website yourself instead of clicking the link. If it claims to be from someone you know, verify through a different channel before doing anything.
And it’s worth thinking about what you keep public. You don’t have to go dark online — but tightening up your privacy settings so strangers can’t see your posts, your school, your location, and your friends list removes a lot of the raw material these attacks depend on.