AI just made scams more convincing — here is what is new and how to defend

For years, one of the most reliable ways to spot a scam email was the typos. Awkward phrasing, odd word choices, grammar that clearly went through a bad translation. That filter is gone. AI writes clean, convincing prose, and attackers are using it.

What has actually changed

Phishing emails that sound like real people. A scam email used to be easy to spot if you read carefully. Today, attackers can feed an AI tool a few samples of someone's writing — scraped from a company website, a LinkedIn post, an email signature — and generate a message that sounds like that person. A fake email from "your accountant" asking you to approve a payment can now match the tone and phrasing of your actual accountant.

Cloned voices on the phone. With a short audio sample — a few seconds from a voicemail, a video, a podcast — AI tools can generate a convincing imitation of someone's voice. This is already being used in real attacks: a call that sounds like the company owner asking someone to approve an urgent wire transfer, or a voice message from "the CEO" asking for a password reset. It sounds wrong because it should sound wrong — but it often does not.

Fake websites and fake chat. AI makes it faster to build realistic-looking sites, generate plausible fake reviews, and staff a chat window with a bot that answers questions convincingly. A supplier verification call that routes to a fake customer-service line, or a payment portal that looks exactly like your vendor's — these are getting harder to identify on appearance alone.

Why the scale matters

The other change is volume. These attacks used to take time to craft, which limited how many a criminal could run at once. AI removes that limit. The same tools that let someone write a hundred convincing personalized emails in an hour also mean that your business is no longer too small to be worth a customized approach. The effort is almost nothing.

The defenses — which are mostly old-fashioned

Here is the part that should be reassuring: the core defenses have not changed much. AI makes the attacks more convincing, but the holes they exploit are the same ones they always were.

Verify money and credential requests on a separate channel. This is the most important one. If you get an email, message, or phone call asking you to transfer money, change payment details, or hand over a password — no matter how legitimate it sounds — pick up the phone and call the person back on a number you already have. Not the number in the email. Not the callback number the caller gave you. A number from your contacts or your records.

This one habit stops the large majority of AI-enhanced financial fraud. The attacker can impersonate the voice, but they cannot intercept a call to a trusted number.

Agree on a code word for urgent finance requests. Some businesses set up a verbal confirmation word — something only internal people know — to verify high-stakes requests over the phone. It sounds unusual until you consider that a cloned voice cannot say a word it does not know.

Slow down on anything marked urgent. Urgency is the mechanism. "Do this right now before it is too late" is designed to short-circuit the instinct to verify. A real emergency can almost always survive a two-minute confirmation call. A scam usually cannot.

MFA on every account, no exceptions. MFA (multi-factor authentication — the secondary code or app prompt after your password) does not stop someone from being tricked into transferring money, but it stops stolen or phished passwords from being used to access accounts. That is still a large category of attack.

The honest frame

AI did not invent fraud. It made some versions of it more polished and easier to run at scale. The underlying mechanics — impersonation, urgency, plausible requests — are the same as they have always been. Your best defenses are also the same: slow down, verify through a trusted channel, and treat any request involving money or access as something worth a thirty-second phone call.

Tenant Strike looks at your Microsoft 365 and Azure setup in read-only mode to find the configuration gaps that make accounts easier to compromise in the first place — things like accounts without MFA, or mail settings that leave your domain open to impersonation. Finding those gaps takes about five minutes.