Is AI Making Phishing Impossible to Spot?

Phishing used to be easy to laugh at: the typo-ridden email from a “bank” you don’t use, the prince with a fortune to share. AI ended the comedy. The scam messages arriving now are fluent, personal, and often reference real details about you — and the instinct that kept a lot of careful people safe (“it looks sketchy, so I’ll ignore it”) no longer works, because looking sketchy was the cheap part and AI made it free to look legitimate.

What actually changed

Phishing has always been a numbers game: send enough bait, some fraction bites. What AI changed isn’t the goal — it’s the economics and the quality of the bait, and that changes how you have to defend yourself.

The grammar filter is gone. For years, the most reliable phishing tell was that the writing was bad. Awkward phrasing and typos filtered out a large share of attacks before you had to think hard. Generative tools produce clean, fluent, professional messages in any language for free. Reading well is no longer any evidence at all that a message is genuine — arguably it’s now neutral, because the scammers upgraded faster than the filter.

Personalization went from expensive to trivial. Convincing, targeted phishing (“spear phishing”) used to take manual research and effort, so it was reserved for high-value targets. AI can now generate personalized messages at scale, weaving in real details scraped from data breaches, social media, and data brokers — your name, your employer, a service you actually use, a recent purchase. A message that knows real things about you feels legitimate in a way a generic blast never did.

The medium expanded beyond email. AI phishing isn’t just text anymore. Voice cloning from a short audio sample powers phone scams where a familiar voice asks for money or access. Convincing fake login pages are cheap to build. Even video can be faked. The attack surface now includes the phone call and the video call, not just the inbox.

The uncomfortable summary: the signals people were trained to look for — how the message reads, how professional it seems — are exactly the signals AI is best at faking. So the defense has to move to signals AI can’t easily fake, which is the rest of this guide.

The tells that still work

Here’s the reassuring part: the content of what a phishing attack needs you to do hasn’t changed, and can’t, because it’s dictated by how the scam makes money. AI can polish the wrapper; it can’t remove the ask. Learn the asks and the polish stops mattering.

• Manufactured urgency. “Your account will be suspended in 24 hours.” “Suspicious sign-in — verify immediately.” “The payment must go out now.” Pressure is the universal phishing ingredient because it short-circuits the pause where you’d otherwise verify. Real organizations rarely demand speed; attackers always do.
• A request for something you should never send. Passwords, full card numbers, one-time codes, or “confirm your details” via a link. No legitimate service asks you to read a security code aloud, ever, or to re-enter your password by following an email link.
• Money that has to move, or a payment method that’s a red flag. Requests to pay, transfer, or buy gift cards — especially urgently — are a scam signature regardless of how polished the pretext is.
• A link or sender that’s almost right. A lookalike domain (support@yourbank-security.com instead of the real domain), or a display name that says one thing over an address that doesn’t match. On a phone, tap or long-press to expand the real address and destination.
• An unexpected code or approval prompt. A two-factor code or “approve this sign-in?” prompt arriving when you didn’t just try to log in means someone else did. Never approve it — and change that password.
• A channel that dodges verification. Pressure to keep the conversation on chat, to not call the official number, to act before you “check with anyone.” Legitimate requests survive verification; scams need to prevent it.

Notice that none of these depend on the message reading badly. They’re about what the message wants — and that’s the part the scammer can’t drop.

The one habit that beats almost everything

If you take a single behavior from this article, take this: when a message or call pressures you about money, access, security, or a code, don’t act on the message itself — verify independently through a channel you already trust.

Concretely:

• Got an alarming “email from your bank”? Don’t click anything in it. Open your banking app or type the bank’s address yourself and check. If something’s genuinely wrong, it’ll be waiting there.
• Got a distressing call from a “family member” or “manager” asking for money or access? Hang up and call them back on a number you already have. A cloned voice can’t answer a callback to the real person.
• Got a “verify your account” link? Reach the site the way you always do — bookmark or typed address — never through the link you were sent.

This works precisely because it sidesteps everything AI is good at. It doesn’t matter how flawless the email reads or how real the voice sounds if you never act inside the channel the attacker controls. The scam depends on you responding to it; the defense is responding around it. It costs you a two-minute pause, and that pause is where nearly every AI-enhanced attack falls apart.

The technical defenses that make phishing much harder

Good habits protect you when you’re alert. The point of the tools below is to protect you when you’re tired, distracted, or briefly fooled — which, given how good the bait is now, will happen to careful people too. Two layers do the heavy lifting.

A password manager — which doubles as a phishing alarm. Beyond giving every account a unique, random password (so one phished credential can’t unlock the rest of your life), a password manager has a quiet superpower against phishing: it autofills your saved login only on the real domain it was saved for. On a lookalike phishing page, it simply won’t offer to fill — because the domain doesn’t match. That silence is a warning worth heeding: if your manager won’t autofill a login you know it has, stop and check the address bar. You just caught a fake the way your own eyes might not have.

Passkeys — phishing-resistant by design. This is the strongest single upgrade available. A passkey signs you in using your device’s fingerprint or face unlock, and crucially there’s no reusable secret to hand over — nothing to type into a fake page, nothing to read to a scammer on the phone. A phishing site literally cannot capture a passkey the way it captures a password or a typed code. Where a service offers to add a passkey, say yes; both password managers above can store them and sync them across your devices. Our afternoon account-security plan covers turning these on step by step.

Keep two-factor authentication on everywhere — but know its limit. Two-factor stops the vast majority of attacks that a leaked password alone would enable, so keep it on. Just be aware it’s not phishing-proof on its own: app and SMS codes can be captured in real time by a fake page and used within seconds, and “prompt-bombing” attacks spam approval requests hoping you tap yes from fatigue. That real-time gap is exactly what passkeys close, which is why they’re the better step where available — and why an unexpected code or prompt should always be refused.

If it already happened: recovery steps

Getting caught by a well-made AI phishing attack is not a character flaw — these are engineered by professionals and the bait has never been better. What matters is acting quickly and in the right order. Do what applies:

If you entered your password on a fake page. Change that password immediately on the real site, and change it anywhere else you used the same one — reuse is what turns a single slip into a cascade. Turn on two-factor authentication if it wasn’t on, and add a passkey while you’re there.

If you approved a login prompt or entered a two-factor code. Assume the attacker got in. Go to the account’s security settings, sign out all active sessions / devices, change the password, and re-check two-factor. Then hunt for what an intruder tends to change: new email forwarding rules, altered recovery phone or backup email, unfamiliar connected apps or devices. Undo anything you didn’t set.

If it was your email account. Prioritize it — email resets everything else. Lock it down first (new password, two-factor, clean recovery details, revoked unknown sessions), then work outward to anything that uses that email for recovery.

If money or a financial account was involved. Contact your bank or card provider right away using the number on your card — not any number from the message — report it, and ask about reversing transfers or reissuing the card. Speed is what limits the loss.

If you shared personal or identity details. Watch for follow-on scams (attackers reuse what they learn), consider a credit freeze if identity documents were exposed, and report the phishing to your email provider and national fraud authority — reporting helps get the operation taken down.

Throughout: stay methodical, not panicked. Phishing relies on rushing you; recovery rewards the opposite.

The bottom line

AI didn’t give phishing a new goal — it gave it a better disguise. Fluent writing, real personal details, cloned voices, and convincing fake pages have retired the “it looks sketchy” instinct for good. But the scam still has to ask you for the same things — money, a password, a code, urgent access — and it still has to stop you from verifying. Move the fight onto ground AI can’t fake: refuse to act inside the message, verify through a channel you trust, let a password manager flag the fake domains you’d miss, and add passkeys so there’s nothing left to steal.

The building blocks live in our afternoon account-security plan and password manager comparison, and the same clear-eyed habit protects you from job scams in the AI era. The full toolkit is at the Privacy & Security Kit hub.