← Back to Cybersecurity
Cybersecurity

AI-Powered Scams: How to Spot the New Generation of Phishing

A phone on a dark desk showing a suspicious message with a distorted voice waveform above it

The traditional advice for spotting a scam — look for bad spelling, odd grammar, a generic greeting — has quietly expired. Generative tools produce fluent messages in any language, clone a voice from a few seconds of audio, and sustain a convincing conversation for as long as it takes. What has not changed is the underlying mechanics: every scam still needs you to act quickly, through an unusual channel, in a way that moves money, credentials, or access. That is where the defence now lives.

What AI actually changed

Three things: volume, polish, and personalisation. Attacks that previously required a skilled operator can now be produced at scale, which means targeted approaches are no longer reserved for large organisations. A convincing message can reference your employer, your recent public posts, your suburb, and a colleague's name, assembled automatically from public sources.

Voice is the most consequential shift. A short clip from a voicemail greeting, a podcast, or a social video is enough to synthesise a familiar voice, and a distressed thirty-second call is hard to evaluate rationally. Video is following: real-time face manipulation on a conference call is no longer a research demonstration, though it still tends to falter under unusual requests, poor lighting, or sudden movement.

The four scenarios worth rehearsing

  • The family emergency call. A familiar voice, urgent trouble, and a request for money or a code, with a reason not to hang up.
  • The executive request. A message or call from a senior colleague asking for an urgent payment, gift cards, or a change to payroll details.
  • The invoice change. A real supplier thread that suddenly contains new bank details and a plausible explanation.
  • The support callback. Someone claiming to be your bank, a platform, or IT, asking you to install a remote access tool or read out a code.

All four share a shape: urgency, authority, an unusual channel, and a request that is irreversible. Name that shape out loud with your family and your team. People resist scams far better when they have rehearsed the specific situation than when they have been told in general terms to be careful.

Defences that still work

Verification out of band is the core technique. Whatever the message says, contact the person or organisation through a number or address you already had, not one supplied in the message. For families, agree a simple phrase or question that would be known only to you, and accept that a genuine relative will not be offended by the check. For businesses, make callback verification a rule for every payment change, applied regardless of who is asking.

Then remove the value of the credential. Passkeys and app-based authentication make stolen codes far less useful, and a password manager will refuse to fill credentials on a lookalike domain — a quiet, reliable signal that a human eye will miss. Set payment limits and delays on accounts where you can, because friction on irreversible actions is the last line that survives when the story is convincing.

Assume that voice and video can be faked, and that fluency proves nothing. The only reliable signals left are the channel, the request, and the urgency — not how the message sounds or reads.

Signals that still give scams away

Look at where the request lands rather than how it is written. A bank that emails, then texts, then calls within an hour is behaving unusually. A colleague who moves from a work system to a personal messaging app is behaving unusually. A payment instruction that arrives late on a Friday, with a reason it cannot wait, is behaving unusually. Time of day, channel changes, and pressure are still the strongest indicators.

Check the technical details calmly: the actual sending address rather than the display name, the destination of a link before you tap it, and whether an attachment asks you to enable anything. On calls, the useful move is not to search for artefacts in the audio but to end the call and ring back on a known number. A real caller will accept that. A synthetic one cannot produce the callback.

What to do in the first hour

If money moved, call the bank immediately and use the words “authorised push payment fraud” or “business email compromise” — speed materially affects recovery. If credentials were entered, change the password from a different device, sign out of all sessions, and check for new forwarding rules, added recovery addresses, or unfamiliar authenticator devices, which is how attackers keep access after a reset.

Then report it. Every country covered here has a national reporting route, and reporting improves the odds of blocking the infrastructure used against the next person. Inside an organisation, make reporting explicitly blame-free. The difference between an incident and a disaster is usually the number of minutes between the click and the phone call.

Regional notes

United States

Report fraud to the FTC and, for business email compromise, to the FBI's Internet Crime Complaint Center as quickly as possible, since rapid reporting can support fund recovery efforts. Contact your bank first; recovery windows are measured in hours, not days.

Canada

The Canadian Anti-Fraud Centre collects reports nationally and publishes current scam patterns worth reviewing with staff. Notify your financial institution immediately and keep records of every message and call for the investigation.

United Kingdom

Suspicious emails can be forwarded to the national reporting address and texts to the standard short code, with Action Fraud handling wider reports. Reimbursement rules for authorised push payment fraud have strengthened, so report to your bank without delay and keep the evidence.

Australia

Scamwatch and the national cyber reporting service collect reports and issue current alerts. Australian banks participate in confirmation-of-payee style checks for some transfers, but these do not cover every case, so out-of-band verification for new payees remains essential.

Questions readers ask

How can I tell if a voice on the phone is AI-generated?

Often you cannot, and trying to judge the audio is the wrong test. Hang up and call back on a number you already had. A genuine caller will accept that; an attacker cannot receive the callback.

Does old phishing advice still apply?

Partly. Spelling and grammar checks are obsolete, but the structural advice holds: verify through a known channel, never act on urgency, and never share codes. Focus on the request and the channel rather than the writing quality.

What is the single best protection for a family?

An agreed verification step for any urgent request involving money or codes — a call back on a known number, or a shared phrase. Discuss it in advance, including with older relatives, so the check feels normal rather than insulting.

What should a business do first after a suspected scam payment?

Call the bank immediately and describe it as business email compromise or authorised push payment fraud, then preserve the emails, check the mailbox for forwarding rules, reset credentials, and report to the national fraud service. Speed is the main variable you control.

The bottom line

AI has made scam messages fluent, personal, and audible in a voice you trust, which retires every defence based on spotting mistakes. What survives is procedural: verify through a channel you already had, put deliberate friction on irreversible payments, use phishing-resistant sign-in, and make reporting fast and blameless. Rehearse the four common scenarios once, and most of this becomes automatic.

This article provides general information, not legal, financial, security, or professional advice. Product terms, availability, privacy practices, and local requirements can change. Check the current information from the relevant provider and seek qualified local advice for high-impact decisions. IntelGuard may earn a commission from future affiliate links, at no additional cost to readers. Read our affiliate disclosure.