Insights

Phishing and deepfake fraud: how a small firm fights back

18 November 2026 · Daniel Haviar

A fraudulent email used to be obvious: broken language, an odd address, a blurry logo. Not any more. AI writes emails that look as if they came from a supplier you know, and a short recording is enough to build a voice that sounds like your director on the phone.

Small offices are exposed. There are few people, everyone opens mail and approves invoices, and a second check feels like a delay. That’s why attackers try you rather than a bank with a fraud desk. You can defend yourself: not by suspecting every email, but with short, agreed rules for payments, account changes and access requests.

What arrives in the inbox now

The most common path is still an attachment or a link. A fake invoice from a known supplier, a “new” account number, a notice that your Microsoft 365 storage is about to run out and you must sign in again. The page the link opens looks like the real one. If you type your password there, and then the code from your phone, the attacker can walk into the mailbox at once.

Once the attacker is inside the mailbox, things often go quiet. They read threads with clients, set a rule that forwards selected messages, and wait for the right moment. Then a message goes out that looks as if the director or the bookkeeper sent it: a change of account number, a request for a quick payment, “I’m travelling, please deal with this now.” A colleague who knows that correspondence has little reason to be suspicious.

A call that sounds like leadership

A deepfake call goes one step further. The screen may show a name you know, and the voice in the earpiece may be one you know as well. Urgency is part of the script: the transfer has to go out within the hour, the notary is waiting, the deal will collapse. The fewer people in the firm who approve payments, the more easily that pressure works.

Technical controls only help so far here. What helps is a rule that applies even to the director: a request for a payment or a change of account is always confirmed on a second channel you agreed in advance. A short chat message is not enough if that chat itself could be faked. You call the number that has been in your contacts for years, not the number in an email you skimmed in a hurry. A second person approving any transfer above an agreed amount is not bureaucracy. It is a safeguard that AI cannot switch off.

What to set in Microsoft 365

People remain the last line of defence, but mail should make their job easier, not harder. SPF, DKIM and DMARC reduce the chance that someone can send messages in the name of your domain. MFA on every account, including leadership, keeps the door closed even after a password is stolen. Conditional Access can block a sign-in from an unexpected place. Mailbox rules deserve an occasional review, because that is where an attacker leaves a back door.

This goes hand in hand with a short, specific training session. Not an hour-long talk about threats, but five examples that look like your ordinary mail, and an agreement on what to do when someone is unsure. People report a suspicious message more readily when they know they will not be blamed for being careful. If an incident does happen, backups and clean permissions decide whether you lose an unpleasant hour or a whole week. There is a separate piece on protecting a small office against ransomware.

If you want to know how your mail, identities and payment rules stand, I will go through them with you. You get a list of concrete gaps, not a general warning that “cyber threats are rising”.

I will review your mail and identities

A security check-up also looks at MFA, mailbox rules, and how easily a fake payment request would get through today.