Your staff are being deceived by AI right now – Why AI security awareness training is not a Tick-box exercise

It is no secret that AI powered attacks are increasing in volume, speed and sophistication. Gartner’s September 2026 AI Security survey identified that 41% of businesses have reported at least one social engineering incident involving a deepfake on an employee audio call in the past 12 months, and 36% reported the same for a video call. Add into this that Verizon’s 2026 Data Breach Investigations Report, showed 62% of all breaches now involve a human element – someone clicking, trusting, transferring, or simply believing what they were shown or told, means there is a gap between employee knowledge of AI powered cyber attacks and what is actually happening from an attack perspective.

So what does this mean? The attack surface has moved forward in terms of speed and complexity. Cyber-attacks are no longer aimed primarily at your organisations firewall, endpoints, or perimeter. Attackers are focusing their efforts at your people, and they are using artificial intelligence to do it with a level of fluency, speed and personalisation that traditional security awareness training was never built to counter. If your organisation’s answer to this shift in attacker behaviour is still a once-a-year
e-learning module and a simulated phishing email with an obvious spelling mistake, you do not have an awareness programme. You have a compliance artefact, and your board should know the difference.

What senior leaders must prioritise right now

Cyber security budgets are finite, and board attention is scarcer still, so prioritisation is not a small point, it is the discipline that determines whether your security spend actually reduces risk. For organisations with a client-facing, trust-dependent business model the priority must shift from pure technical control to human resilience, and it has to shift now.

That means three things rising to the top of the agenda. First, deepfake recognition, specifically, the ability of staff who handle payments, approvals, or sensitive instructions to recognise and challenge a synthetic voice or video, even when it sounds exactly like their chief executive. Second, recognising that AI-driven phishing – email, SMS and collaboration-platform messages are no longer generic and clumsy, but individually crafted using scraped data about your organisation, your clients and your internal relationships. Third, taking the time to understand behavioural social engineering – the multi-channel, multi-step campaigns that combine a believable pretext with urgency and authority, exploiting workplace norms rather than technical vulnerabilities.

This is not a call to abandon technical controls, email filtering, endpoint detection and identity controls remain essential. It is a call to recognise that technical controls alone cannot stop an attack that is designed, from the outset, to be convincing to a human being. Your focus must widen to include the one layer of your defence that AI has made dramatically harder to trust. Your own judgement, and your people’s.

The situation organisations are actually facing

Look honestly at what has changed in the past year, and the picture is painful. The barrier to entry for convincing impersonation has collapsed. Cloning an executive’s voice from a handful of public clips, or generating a passable video likeness, no longer requires specialist skill or significant cost, it requires motivation and a few minutes with freely available tools. This isn’t just hearsay or scaremongering. The FBI’s Internet Crime Complaint Centre logged its first standalone AI-fraud category in 2025, recording 22,364 complaints and $893.3 million in adjusted losses. A figure that is itself likely an undercount, since many incidents are never reported or never recognised as AI-enabled in the first place.

At the same time, the economics of social engineering have shifted sharply in the attacker’s favour. AI-generated phishing content is cheaper, faster and more targeted to produce than ever before, which means your organisation is not facing occasional, crude attempts. It is facing a sustained, scaling campaign that specifically studies how your business operates before it strikes. Attackers now research your supplier relationships, approval chains, and internal language, then build a pretext that fits precisely inside your normal way of working. That is why so many of these incidents succeed not because an employee was careless, but because the deception was engineered to look exactly like a legitimate, everyday request.

The organisations most exposed are not necessarily the ones with the weakest technology. They are the ones whose culture rewards speed and responsiveness over verification. A client-focused ethos, ironically, can become a vulnerability if staff are never given explicit permission, and practical tools, to pause and challenge an instruction that feels slightly wrong. A finance team trained to be helpful and fast is precisely the team a well-crafted deepfake call is designed to exploit. This is not a failure of character. It is a failure of preparation, and it sits squarely with leadership to correct it.

What a practical, proportionate response looks like

The answer is not a bigger slideshow or more generic security email training. It is a bespoke, continuously evolving training programme built around the specific threat vectors your organisation actually faces, delivered to employees and key stakeholders in a form they will retain under pressure. Because the moment that matters is not during the training session, it is during the call, the email, or the video request itself.

From our experience at Cyberfort, having delivered this type of training to several organisations in recent weeks, we have seen how AI security training requires a genuinely practical programme that does four things which generic, off-the-shelf awareness training does not. It teaches deepfake recognition as a specific, rehearsed skill – what synthetic audio and video actually look and sound like, what verification step to take when something feels urgent or unusual, and crucially, that staff have explicit organisational permission to slow down and check, even when the voice on the line sounds exactly like the finance director for example.

A good AI cyber training programme should build fluency in AI-driven phishing patterns, not spotting typos, but recognising the signs of content that is individually tailored, contextually accurate and emotionally engineered to bypass scepticism. It trains behavioural social engineering awareness across every channel your business uses, not just email, because today’s campaigns move fluidly between email, SMS, collaboration platforms and phone calls within a single attempt. And it should align every one of these elements with your organisation’s own risk-aware culture and client-facing ethos, so the training reinforces how your people already think about trust, rather than fighting against it.

This has to be measured, not assumed. A credible programme includes simulated AI-threat scenarios – realistic, evolving tests rather than static exercises, and clear reporting pathways so that staff who spot something suspicious know exactly where to raise it and feel confident doing so without blame. The organisations seeing the strongest improvement in resilience are the ones treating this as an ongoing capability-building exercise, refreshed as attacker techniques evolve, not a once-a-year box to tick before the next audit.

Why AI security training needs specialist MSSP support

For any organisation considering building this capability in-house it is highly likely that your internal IT or security team almost certainly does not have the specialist expertise, attacker-informed intelligence, or the capacity to design and continuously update AI-threat awareness training to the standard your organisation demands.

This is not a criticism of any IT or Security team. It reflects how fast the threat landscape has moved and how specialised effective countermeasures have become. Deepfake detection, AI-phishing pattern recognition and behavioural social engineering defence are now disciplines in their own right, informed by live threat intelligence that a managed security partner sees across many organisations, not just yours.

A specialist managed security services provider brings three things an internal team struggles to replicate alone when it comes to AI security training:

The organisations that treat security awareness as a strategic capability, built and maintained in partnership with a specialist that lives in this threat landscape every day, are the ones that will still be trusted by their clients and their boards when a well-crafted deepfake call eventually reaches their teams. The organisations that treat it as a compliance exercise will find out the hard way that a tick-box training module was never going to stop an attacker who had already done their homework.

The next step is not complicated. It is a conversation about where your people are most exposed. It should cover what your current training actually covers against the threats of 2026 rather than 2022, and what a bespoke programme aligned to your culture and your risk profile would look like in practice. That conversation is worth having before the incident, not after it.