Photo by Berkeley Communications on Unsplash
A member of your finance team gets a phone call. The voice on the line sounds exactly like your managing director. The caller explains there is an urgent payment needed before close of business. The tone is calm, the context makes sense, and nothing about the request feels out of place.
Except the managing director never made that call. The voice was generated using AI, trained on audio pulled from a company video, a podcast appearance, or a recorded Teams meeting. By the time anyone thinks to check, the transfer has gone through.
This is not a theoretical risk. It is happening to businesses of every size, including small ones, and the tools making it possible are cheap, widely available, and getting better every month.
Why the Old Warning Signs Are No Longer Enough
The standard advice for years was to look for bad grammar, strange sender addresses, and awkward phrasing. Those signals still matter, but they are no longer reliable on their own. As we covered previously, spotting bad grammar is no longer a security strategy for UK small businesses. AI tools can now produce fluent, contextually accurate text in seconds, personalised with real details drawn from your website, your LinkedIn page, or your staff directory.
Attackers no longer need to be clever. They need a free tool and a few minutes of preparation.
What AI-Powered Scams Actually Look Like Right Now
It helps to understand the specific forms these attacks take, because they each require a slightly different response:
- Voice cloning calls. AI can replicate someone's voice from a small amount of recorded audio. Attackers use cloned voices to call staff members with urgent requests, most commonly involving payments, credential resets, or access to sensitive systems.
- Hyper-personalised phishing emails. Rather than sending generic account warnings, attackers now craft messages that reference your business name, the recipient's actual role, a recent event, and the name of a real colleague. The result reads like genuine internal correspondence.
- Deepfake video messages. Short video clips of a known person, appearing to share sensitive information or request urgent action, can be generated from publicly available footage. This is less common at the small business level today, but the capability is becoming more accessible.
- AI-driven conversation impersonation. Automated systems can now hold convincing text-based conversations over time, posing as IT support, a supplier, or HR, gradually extracting credentials or payment details rather than making one obvious ask.
Why Small Businesses Are in the Frame
There is a persistent assumption that attackers focus on large organisations. The reality is that small businesses are often more exposed precisely because they tend to have less formal processes around payment authorisation and data requests. A single well-timed call to the right person can be enough. The barrier to running these campaigns has also dropped sharply. We looked at this in detail when examining how AI is making cybercrime easier to launch. Attackers can now run personalised, targeted campaigns across dozens of small businesses simultaneously at very low cost.
What Actually Helps
Technology alone will not stop a convincing voice cloning call. The most effective defence is a combination of clear human habits and simple process controls.
- Verify through a second channel before acting. Any unexpected request involving a payment, a credential change, or access to sensitive data should be confirmed via a separate contact method. Call the person back on a number you already hold. Never use contact details supplied in the message itself.
- Set up a simple confirmation step for high-risk requests. Some businesses use a short verbal or written codeword for payment requests above a certain value, something only the real person would know. It is low-tech, takes five minutes to implement, and removes a significant amount of the risk.
- Build the habit of pausing, not just recognising. The goal is not to make every employee a security analyst. It is to create a reliable pause before acting on any urgent financial or access request that arrives unexpectedly. Security awareness training designed around real scenarios is one of the most practical ways to build that habit across your whole team.
- Keep devices and software updated. AI scams frequently rely on follow-up malware or credential harvesting tools that exploit unpatched systems. Keeping everything current limits the damage if someone does respond to something they should not have.
- Be selective about what you publish publicly. Voice cloning needs audio. Deepfakes need video. Personalised phishing needs context. A business that posts extensive director video content, detailed staff listings, and granular operational information gives attackers considerably more to work with.
A Short Practical Takeaway
AI scams work because they are designed to bypass the instincts your team has spent years developing. The request feels normal. The voice sounds right. The email reads like something a genuine colleague would send. That familiarity is the whole point.
The response is not blanket suspicion of everything, which is exhausting and gets in the way of real work. It is a small number of specific habits applied to a narrow set of high-risk actions: payments, credential changes, and requests for access. Those habits, consistently followed, take away most of what these attacks depend on.
If your team has not yet had a structured conversation about AI-generated scams, that is the most useful place to start. Structured awareness training can turn that conversation into something your people will actually remember and use.