A finance manager receives a Teams message that appears to be from the managing director: a supplier’s bank details have changed, the payment is urgent, and a quick voice note confirms it. The wording is clear, the tone sounds familiar and the request fits a busy day. That is the risk posed by AI phishing. It gives criminals a faster, cheaper way to create convincing impersonation attempts that can slip past people who already know the usual warning signs.
For UK small and medium-sized businesses, this is not simply an email security issue. A successful attack can lead to fraudulent payments, exposed Microsoft 365 accounts, interrupted operations, client confidentiality concerns and difficult conversations with insurers, auditors or regulators. The practical response is to make fraud harder to execute, even when a message looks and sounds credible.
What makes AI phishing different?
Traditional phishing messages often gave themselves away. They contained obvious spelling errors, awkward phrasing, generic greetings or implausible requests. Those clues still matter, but they are no longer reliable on their own. Generative AI can produce polished, natural English in seconds, tailored to a particular industry, seniority level or business relationship.
Criminals can also use public information from company websites, social media profiles, tender notices and data breaches to make a message more believable. They may know who works in finance, which software a firm uses, where a director is travelling or the names of regular suppliers. AI helps turn those fragments into a convincing story at scale.
The result is often not a dramatic scam but a plausible request: review a shared document, re-authenticate Microsoft 365, approve an invoice, buy gift cards for a client event, or change payroll details. It arrives when the recipient is busy and is designed to encourage speed over checking.
The threat now reaches beyond email
Email remains the main route into many organisations, but it is not the only one. AI-assisted phishing can appear through Teams, WhatsApp, LinkedIn, text messages and fake support calls. A criminal may use a realistic email first, then follow up with a call or voice note that creates urgency.
Voice cloning is a particular concern. A short sample from a public video, podcast or voicemail can sometimes be enough to create an imitation. It may not withstand careful scrutiny, but a rushed employee who believes they are speaking to a director may not pause long enough to question it.
This does not mean every unusual message is AI-generated, nor that businesses should distrust every colleague. It means the old test of “does this look professionally written?” is no longer sufficient.
How AI phishing creates business risk
The immediate loss from a fraudulent payment is often the most visible consequence. However, an attacker who captures a user’s Microsoft 365 login can do far more. They may search mailboxes for invoices and bank details, set up forwarding rules, impersonate the account internally, or wait quietly for an opportunity to intercept a payment.
For professional services firms, an exposed mailbox can reveal client correspondence, personal information and commercially sensitive documents. For logistics and operational businesses, it can disrupt supplier communications, delivery arrangements and access to systems needed to keep work moving. In regulated sectors, the incident may also trigger reporting and evidence-gathering obligations.
There is a human cost too. Staff who made an honest mistake can feel embarrassed or worried about their job. A good security culture treats reporting as the right action, not an admission of failure. Early reporting gives the business a better chance of stopping account takeover or recalling a payment.
The checks that still work against AI phishing
The most effective defence is not asking staff to become forensic investigators. It is establishing a small number of clear habits that apply whenever money, credentials, personal data or urgency are involved.
First, verify requests using a separate, trusted channel. If an email asks to change bank details, call the supplier using a number already held in your records, not one in the message. If a senior colleague sends an unusual Teams request, call them on their known number or speak to them directly. Do not reply to the potentially compromised message to confirm it.
Second, make payment changes a process rather than a judgement call. A documented callback procedure, dual approval for relevant payments and clear supplier verification steps create useful friction. They may add a few minutes, but that is a sensible trade-off against the cost of a misdirected payment. The process should apply equally when the request appears to come from the owner, a long-standing client or a trusted supplier.
Third, pause before entering credentials. Microsoft 365 phishing pages can closely resemble the genuine sign-in screen. Staff should be especially cautious if they arrive there through an unexpected link, a shared-file notification or an urgent security warning. Opening Microsoft 365 through the usual bookmark or app, rather than from the message, is a simple safer habit.
Finally, report anything doubtful quickly. A suspicious message sent to the right internal contact can help protect everyone else. It also allows technical teams to check whether similar emails were delivered, whether anyone clicked, and whether accounts show signs of unauthorised access.
Technical controls that reduce the pressure on staff
Training matters, but no person will identify every well-crafted scam, particularly during a pressured working day. Technical controls should reduce the number of malicious messages that arrive and limit what an attacker can do if one gets through.
Email protection should assess links, attachments, sender identity and impersonation signals before messages reach inboxes. Domain protections such as SPF, DKIM and DMARC help reduce spoofing of your own organisation’s email domain. They are not a complete answer to lookalike domains or compromised supplier accounts, but they remain essential foundations.
Multi-factor authentication is equally important, preferably using phishing-resistant methods where appropriate. A stolen password should not be enough for a criminal to access email, cloud files or remote systems. Conditional access policies, sensible sign-in controls and prompt removal of unused accounts further reduce exposure.
It is also worth reviewing who can approve payments, create mailbox forwarding rules, register new devices and access sensitive shared folders. Least-privilege access will not stop every phishing attempt, but it limits the damage when an account is compromised. For many businesses, this is where managed oversight takes pressure away from internal teams that are already busy running the operation.
Backups are part of the picture as well. Phishing is often the first stage of ransomware or data theft. Tested backups for servers, files and Microsoft 365 data give the business more options if an attacker deletes, encrypts or alters information.
Build a response plan before an incident
When an employee suspects they have clicked a malicious link or disclosed a password, the first hour matters. Staff should know exactly who to contact and should be encouraged to act immediately, even if they are unsure. Delay is more damaging than a false alarm.
A practical incident procedure should cover who can reset passwords and revoke sessions, how to isolate a device where needed, who contacts the bank after suspected payment fraud, and how the business will communicate with affected clients or suppliers. It should also identify decision-makers who can be reached outside normal hours.
Testing the process is worthwhile. A short phishing awareness exercise can show whether people know where to report messages, while a tabletop scenario can reveal gaps in payment approvals or account recovery. The aim is not to catch staff out. It is to make the right response familiar before a real event creates pressure.
AI phishing will continue to improve because it is profitable, scalable and easy for criminals to adapt. The answer is not fear, and it is not expecting employees to spot every trick. Clear verification rules, well-managed Microsoft 365 security, layered email protection and supportive training give your team something far more useful: the confidence to stop, check and carry on with their work safely.