In this guide · 8 sections
- 01What official reports say about AI in phishing
- 02The old red flags vs the new ones
- 03Process red flags that still work
- 04Voice cloning, deepfakes and QR codes
- 05Technical controls that do the heavy lifting
- 06An updated phishing checklist for staff
- 07Do it yourself or get help?
- 08Frequently asked questions
Short answer
AI-written phishing is fluent and personalised, so judge the request, not the writing. Treat urgent payments, bank-detail changes, login requests or moves to a new channel as suspicious, and verify via a known contact. Back it up with MFA, email filtering and phone checks on payment changes.
What official reports say about AI in phishing
CSA's Singapore Cyber Landscape 2025/2026, published 30 June 2026, names the spread of artificial intelligence as a key driver of the year's threats. On phishing specifically, it says:
- Phishing emails remained the main social engineering method, using AI-generated lures that are authentic-looking and contextually relevant.
- Attackers added AI-enabled voice cloning and video deepfakes.
- Toolkits were built to bypass MFA using adversary-in-the-middle tools and QR-code phishing ("quishing"), used to get victims to authorise payments or change account details.
- Around 4,800 phishing attempts were reported to CSA in 2025, down from 6,100 in 2024, but CSA says reported numbers likely understate the true scale.
An industry chapter in the same report adds that attackers use generative AI to produce localised phishing in multiple languages, which removes the need for native speakers. In other words, a fluent message in your customers' language is no longer evidence that it is genuine.
The old red flags vs the new ones
Most phishing training still lists clues from a time when scam emails were written quickly by non-native speakers. Some still help, but the reliable signals have moved from how a message looks to what it asks you to do.
| Old red flag | Still reliable? | Use this instead |
|---|---|---|
| Spelling and grammar mistakes | No: AI writes fluently | Does the request make sense for this sender? |
| Generic "Dear customer" | No: AI personalises easily | Did you expect this message at all? |
| Padlock or HTTPS on the link | No: most phishing sites use HTTPS | Did you type the address or use a bookmark? |
| Strange sender address | Partly: look-alike domains are common | Check the full address, then verify by a known contact |
| Urgency and pressure | Yes | Any pressure to act now is a reason to slow down |
| Request for money, codes or logins | Yes | Always verify through a separate, known channel |
CSA's report backs up the HTTPS point: 79% of phishing websites reported to CSA in 2025 used HTTPS, up from 69% the year before, and CSA says users should judge legitimacy on factors beyond the encrypted connection.
Process red flags that still work
Fluent writing can be faked. Your business processes cannot, unless someone agrees to skip them. These are the signals to teach.
Stop and verify if you see any of these
The response is the same for all of them: stop, and confirm through a contact you already have, such as the phone number on file or a message to the person's known account. Never use the phone number, link or reply address in the message itself.
Voice cloning, deepfakes and QR codes
CSA's report lists voice cloning and video deepfakes alongside email. For a small business, the realistic version is a call or voice note that sounds like the owner or a regular supplier, asking for an urgent payment or a change of details.
QR-code phishing works because a QR code hides the link. A QR code in an email or on a sticker that leads to a login page should be treated like any unexpected link: close it and go to the site the usual way.
Technical controls that do the heavy lifting
Training helps people pause, but technical controls catch what slips through. Most of these are already in Microsoft 365 or Google Workspace business plans.
Turn on MFA everywhere, phishing-resistant for admins
MFA stops most stolen-password logins. CSA notes kits that bypass weaker MFA, so use passkeys or security keys for admin and finance accounts where you can.
Tune your email filtering
Turn on safe links, attachment scanning and impersonation protection for directors and finance staff in Microsoft Defender for Office 365 or Google Workspace.
Add an external sender banner
A visible tag on emails from outside the company makes a fake "internal" message stand out.
Publish SPF, DKIM and DMARC
These make it harder for attackers to send email that pretends to come from your own domain.
Require two people for payments and bank changes
Set maker-checker approval in business banking so one tricked person cannot send money alone.
Give staff a report button
A one-click way to report suspicious mail means the first person to spot a campaign protects everyone else.
If you are working toward CSA's Cyber Essentials mark, these controls overlap with its requirements. Our Cyber Essentials checklist maps them out.
An updated phishing checklist for staff
Keep it short enough to pin next to a screen. The old checklist asked "does this look fake?". The new one asks "what is this asking me to do?".
- Was I expecting this? If not, be suspicious however professional it looks.
- What does it want? Money, a login, a code, a file, or a change of details means verify first.
- Is it rushing me? Urgency is the attacker's tool. Slow down.
- Is it moving me to a new channel? A switch to WhatsApp, a personal email or a phone number in the message is a warning sign.
- How do I check? Use a contact you already have, or type the website address yourself.
- Who do I tell? Report it, even if you already clicked. Fast reporting limits the damage.
Our guide to spotting a phishing email covers the technical clues in more detail, such as checking the real sender address and link destinations. For staff who use AI tools at work, CSA and IMDA's joint advisory on safe use of generative AI (28 July 2026) also advises verifying AI-generated content before using or sharing it.
Do it yourself or get help?
Two changes cost nothing and work today: a written call-back rule for payments and bank changes, and a five-minute team briefing on the process red flags above. Turning on MFA and the external sender banner is usually an hour's work for whoever administers your email.
If you want it done properly, our Security Awareness Training (S$390) sets up a 12-month plan of five-minute lessons with Singapore scam topics and completion tracking, in your own account. Email Threat Protection (S$290) tunes Microsoft 365 or Google Workspace protection, impersonation rules for directors and finance staff, the external banner and a report-phishing button. If you think a scam already reached your team, the ScamShield site lists what to do next.
Frequently asked questions
Not reliably by how they read. AI produces fluent, personalised text, so spelling and tone are poor guides. Email filters still catch many by sender reputation, links and attachments. For people, the dependable test is the request: urgent payments, bank changes, login or code requests and moves to a new channel should always be verified separately.
No. The padlock only means the connection is encrypted, not that the site is genuine. CSA reported that 79 percent of phishing websites reported to it in 2025 used HTTPS. Reach important sites by typing the address or using a bookmark, not by following links in messages.
Agree a rule that no payment, bank-detail change or password reset is ever carried out on the strength of a call, voice note or video alone. The person receiving the request hangs up and calls back on a number already saved. Combine this with two-person approval for payments in your business banking.
It stops many attacks that rely on a stolen password, which AI-written phishing makes easier to collect. CSA notes toolkits that bypass some MFA methods, so use phishing-resistant options such as passkeys or security keys for administrator and finance accounts where possible.
Yes, if it is updated. Training that focuses on spelling mistakes is out of date. Short, regular lessons about process red flags, such as payment changes, urgency and requests to switch channels, plus an easy way to report suspicious messages, remain one of the cheapest defences for a small business.