AI email security
AI writes better phishing than most people write email. Filters and process have to keep up.
The spelling mistakes are gone. AI-written messages use the right names, the right projects and the right tone, and some carry no link at all. Machine learning filtering catches a lot of it. The rest is stopped by process. Here is how the layers fit, and a map of which attacks your current defences cover.
Claims data from Coalition; Microsoft licensing from Microsoft Learn. Checked 31 October 2026.
- How AI phishing works
- What filters catch
- Coverage map
Reviewed by Blake Bath, General Manager, on 31 October 2026.
AI email security
- Credential phishing
- Impersonation
- Compromised supplier
- Malicious attachments
Read this page as
You want the business to keep running when something breaks, and to stop being the person who has to sort it out.
Start withWhat arrivesCoverage mapKinds of AI
People come to you when the email, the printer or a laptop stops. You want one number to call and someone who owns it until it is fixed.
Start withHow it worksWhat arrivesQuestions
You want to know what is included, what it costs over a year and what you will have in writing for the insurer and the auditor.
Start withCoverage mapThe layersQuestions
Time and money are tight, most of your people are not IT people, and your board asks about risk.
Start withWhat arrivesHow it worksKinds of AI
What we see
Four kinds of AI-assisted email attack.
Each needs a different layer to stop it.
Credential phishing
A polished message with a link to a fake sign-in page.
- Link checking
- Phishing-resistant MFA
- Report button
Impersonation
A lookalike domain or display name posing as a leader or supplier.
- Impersonation rules
- External tags
- DMARC
Compromised supplier
A real supplier's mailbox, used to send a fake invoice or bank change.
- Behaviour analysis
- Phone check
- Two approvers
Malicious attachments
Documents that run code or lead to a download.
- Sandboxing
- Macro blocking
- EDR
Want the whole picture?See how it worksTalk to us
Email defence coverage map
Which email attacks do your defences actually cover?
Tick the layers you have. You see which attack types each helps with and the gaps, in the order we would close them.
Tick the layers you have today. You see which email attacks each one helps with, and where you are exposed.
Your coverage
2 of 7 attack types
- Covered: Someone sending as your own domain
- Gap: Lookalike domains and display-name tricks
- Gap: Links to fake sign-in pages
- Covered: Malicious attachments
- Gap: A real supplier's mailbox, compromised
- Gap: A clean, AI-written payment request with no link
- Gap: An attacker already inside a mailbox
Next, in order
- Phone verification on every bank detail change. Free, and it stops the attacks filters miss.
- Add a filtering layer that rewrites links and checks attachments and impersonation: Defender for Office 365 or a dedicated email security product.
- Phishing-resistant MFA, so a stolen password from a fake page isn't enough.
- Alerts on new forwarding and delete rules, acted on quickly.
- A report button, and a fast answer when staff use it.
Covered means at least one layer helps. No single layer stops everything.
Want a person to look at your answer?Talk to usBook a free 30 minute chat (opens in a new tab)
Be specific
Not all AI is the same.
- Generative AIWrites the message. Used by attackers to produce convincing text at scale.
- Machine learningReads patterns across huge volumes of mail. Used by defenders to score new messages.
- PeopleBring context: who is on leave, which supplier is in dispute, what just feels wrong.
"Real Bytes run the entire IT department for my business. The service is nearly instant, and proactive with advice. Outstanding."
Defence in depth
The layers, and what each one does.
| Layer | What it does | Where it comes from |
|---|---|---|
| DMARC at reject | Stops others sending as your domain | DNS settings, any plan |
| Built-in filtering | Spam, known malware, bulk mail | All Microsoft 365 and Google plans |
| Defender for Office 365 Plan 1 | Link and attachment checking, impersonation protection | Business Premium; Microsoft 365 E3 from 1 July 2026 |
| Dedicated email security layer | Extra analysis, including compromised supplier mail | Separate product, such as Check Point Harmony Email & Collaboration |
| Phishing-resistant MFA | Makes a stolen password from a fake page useless | Entra ID and passkeys |
| Process | Phone checks on bank changes, two approvers | Your procedures |
Not sure where you stand?Get a free checkTalk to us
How it chains
A typical attack, step by step.
- 1. A message from a known supplierOpened on a phone, between meetings.
- 2. A link to a sign-in pageOn a domain registered days ago.
- 3. A mailbox rule appearsForwarding finance threads to an outside address.
- 4. A bank change request followsWeeks later, perfectly written, in the real thread.
Each step on its own looks ordinary. Filtering, MFA, mailbox rule alerts and a phone check each break the chain at a different point.
Our view
Buy filtering for volume. Rely on process for money.
Good filtering removes most of the noise, and machine learning is what makes that possible at scale. But the most expensive emails we see are the clean ones: a real supplier's account, a normal-looking thread, a new bank account.
So we set up the layers in order: DMARC, phishing-resistant MFA, Defender presets or a dedicated layer, mailbox rule alerts, and a phone check on every bank detail change. That last one costs nothing.
The order
Strengthening email defence in five steps.
- Step 1DMARC to reject
No one sends as you.
- Step 2Phishing-resistant MFA
Stolen passwords stop working.
- Step 3Filtering presets
Defender for Office or a dedicated layer.
- Step 4Alerts and reporting
Mailbox rules watched, report button used.
- Step 5Payment process
Phone checks and two approvers.
Ready when you are.Book a free 30 minute chat (opens in a new tab)Talk to us
Related guides
Email security in depth. Filters, sign-in and process.
The controls around the inbox.
Questions
What people ask about AI and email security.
How does AI change phishing?
It removes the old warning signs. Messages are well written, use the right names and projects, and can be produced in volume and tailored to each target. Some carry no link or attachment at all, just a convincing request.
What is the difference between generative AI and machine learning in email security?
Generative AI writes text and images, and attackers use it to write phishing. Machine learning spots patterns across huge numbers of messages, and defenders use it to score new mail against what normal looks like.
Is Microsoft 365's built-in filtering enough?
Basic filtering handles spam and known malware well. Defender for Office 365, in Business Premium and in Microsoft 365 E3 from 1 July 2026, adds link and attachment checking and impersonation protection. A dedicated email security layer adds more, especially for compromised supplier mailboxes. What you need depends on your risk.
Can a filter stop a clean AI-written payment request?
Not reliably. A message with no link, from a real but compromised supplier, may look completely normal. That is why phone verification of bank detail changes is the control that matters most.
Can AI replace the people reviewing email threats?
No. Machine learning is good at pattern and poor at context. It doesn't know your CFO is on leave or that a supplier has just changed banks. People still make the final call on the odd ones.
Where should a small business start?
DMARC at reject, phishing-resistant MFA, Defender for Office 365 preset policies if you have them, a report button staff actually use, and a phone check on every bank detail change. Then consider a dedicated email security layer.
Next step
Want to know what's getting through?
We review your filtering, DMARC, sign-in and payment process, and show you which attacks your current setup would miss.
What happens next
- You pick a time that suits you.
- One of our engineers calls you. It takes 30 minutes and it's free. There's no slide deck.
- If we can help, we send a written plan. If we can't, we'll tell you who can.
- Nothing happens until you say yes.
You talk to the people who do the work, not a sales team. Rather talk now? Call 07 3114 2808.
Already a client? Log a job in the client portal (opens in a new tab)
Keep reading
- Email security gateway Our service.
- Email security DMARC and more.
- BEC Guide.
- Pricing Plans.
- IT support Brisbane Published prices and response times, with engineers who come on site.
- IT services for Brisbane businesses What we do and how we work.
Reviewed by Blake Bath, General Manager, on 31 October 2026. Claims data from Coalition; Microsoft licensing from Microsoft Learn. Check Point and Harmony Email & Collaboration are trademarks of Check Point Software Technologies Ltd. Checked 31 October 2026. Coalition, 2025 Cyber Claims Report; Microsoft Learn, Defender for Office 365 service description; ASD, Protecting against business email compromise

Remote