AI-Written Phishing Emails Are Fooling Half the People Who Get Them

AI-Written Phishing Emails Are Fooling Half the People Who Get Them. Here’s What Small Businesses Need to Change

Phishing used to be easy to spot. Bad grammar, a weird sender address, a link that clearly went somewhere sketchy. Most people learned to catch it without thinking twice.

That’s over now. AI-crafted phishing emails are getting click-through rates around 54%, compared to roughly 12% for the traditional versions. That’s not a small improvement for attackers. That’s a different game entirely, and small businesses are squarely in the blast radius.

Why the old training doesn’t work anymore

The advice you probably gave your team a few years ago, watch for typos, check the sender, hover over links, still matters. But it’s no longer enough. AI tools can now write an email that sounds exactly like your vendor, your bank, or your boss. It can mimic tone, reference real details, and skip every red flag your team was trained to look for.

And it’s not just email. Voice cloning and deepfake video are cheap and accessible now. A convincing fake voicemail from “the owner” asking someone to wire funds or share login credentials isn’t science fiction. It’s a Tuesday.

The other half of the problem: your own team

Here’s a stat that doesn’t get talked about enough. Nearly half of employees still use personal AI accounts for work tasks, pasting client information, financial data, or internal documents into tools that were never approved or secured by the business. That’s not malicious. It’s just people trying to get their job done faster. But every one of those shortcuts is a door nobody’s watching.

Small businesses often respond to this by banning AI tools outright. That doesn’t work. It just pushes the behavior underground, where you have even less visibility into what’s happening with your data.

What actually helps

Give people an alternative, not a lecture. If employees are reaching for AI tools to save time, provide an approved option that’s actually good, so they’re not tempted to go find their own.

Put a second set of eyes on money and access. Any request to move funds, change payment details, or share credentials should require a second human to confirm it, through a different channel than the one the request came in on. If someone gets a suspicious wire request, a phone call to a known number settles it in thirty seconds.

Train your team on the new reality, not the old one. Deepfakes and cloned voices are the current threat, not a future one. A quick internal briefing on what these attacks look like now goes further than another slideshow about spotting typos.

Write down how AI is used in your business. It doesn’t need to be complicated. A short document covering what tools are approved, what data can and can’t go into them, and who’s responsible for reviewing it does two things: it protects you operationally, and it matters when you’re talking to an insurer. Coverage interpretations on AI-related incidents vary a lot right now, and demonstrated governance is what gets you better terms.

Check your policy. Cyber and professional liability coverage doesn’t automatically include AI-related incidents. Ask your broker directly whether your current policy covers a loss caused by an AI-generated attack, and get the answer in writing.

The takeaway

None of this requires a security department or a six-figure consulting engagement. The National Institute of Standards and Technology publishes a free AI Risk Management Framework built for exactly this, practical guidance without the enterprise price tag.

The businesses getting hurt by this shift aren’t the ones using AI. They’re the ones who never stopped to look at where the actual gaps were before something exploited them. That’s the diagnostic step most companies skip, and it’s usually the cheapest one on the list.