In 2019, BBC News invited me to appear on television to talk about the startling growth of cybercrime. My message was stark: fraud had become a pandemic. Seven years later, Guildhawk have taken that same interview and applied the AI dubbing they use for compliance training to make both the BBC interviewer and me speak Mandarin.

2019: Original BBC cybercrime interview                                                  2026: Same interview now in Mandarin

In 2019, doing that convincingly with amazing lip sync would have sounded like sci-fi. Today, AI dubbing and digital human avatar solutions can make it possible to take a piece of training created in one language and scale it across your global workforce.

And that raises a much more interesting question.

What will be possible seven years from now?

For compliance, safety and learning professionals, I think the answer matters enormously.

The breakthrough isn't really a better-looking avatar – the very first one that gave my chairman’s address for the Fraud Advisory Panel during lockdown looks so dated now. No, it is the prospect of giving people around the world consistent, accurate information quickly in their own language, without having to recreate the entire learning experience country by country.

Clients like Barclays, Sandvik Canada and the police already entrust Guildhawk to transform their multilingual training with a combination of AI dubbing, digital humans, expert linguistic verification and e-learning production. Our current model is deliberately hybrid: subject matter experts verify translated scripts and approved terminology, Guildhawk produces and checks the video, and the resulting content with SCORM files can be delivered straight into the client's LMS.

        Barclays customer information provided in Hindi                                  The same in English 

But buying an avatar isn't a future-proofing strategy. Building the right foundations for the new AI Age is.

LEARN MORE: How AI avatars and dubbing are used for compliance training

Here are three things I believe compliance and safety leaders should be thinking about now.

1. Treat your knowledge as an IP asset, not AI fuel

Everyone says data is the fuel that powers AI, pollute it and guess what it does to the engine? So, for organisations in regulated industries, I would put it slightly differently:

Trusted data is the fuel of trusted AI.

Your approved terminology, safety procedures, regulatory reports, translations, HR policies and training scripts represent years of institutional knowledge.

Before feeding that material into AI models, organisations should be asking some basic questions. Where is our proprietary knowledge going? Who can access it? How is it protected? And are we comfortable with the contractual arrangements governing its use?

There is another consideration that is easily overlooked. A translation isn't simply "content in another language". In safety and compliance training, a badly translated instruction, legal definition or technical term can change the meaning of what the employee is being taught. If that weakness is discovered by an inspector or insurer after a serious incident, and found to be relevant, it could be a big issue.

That is why Guildhawk's approach to creating multilingual multimedia production with or without AI, always includes verification of outputs by an expert linguist. Be it a short dubbed interview of the CEO announcing annual results or scaling 8 anti-bribery modules into 30 languages with avatars, the ironclad safeguards apply. This means ISO:27001 certified security controls for protecting IP and ISO 9001-backed expert review of translation accuracy.

This is also why the idea behind Guildhawk's Avatar SafeHouse™ matters. Custom avatars are managed under Guildhawk's ISO:27001 controls rather than treating someone's digital likeness as just another disposable media file you see on social media.

Future-proofing principle No. 1: protect your knowledge, curate it and keep it accurate.

Because tomorrow's AI is only as trustworthy as the knowledge you allow it to work from.

2. Prepare for avatars that don't just present, but converse

Today's AI avatar technology is remarkable.

Tomorrow's will be even more powerful.

We are moving from AI that primarily generates or retrieves information towards systems capable of increasingly sophisticated reasoning, inference and interaction. But reasoning does not magically eliminate error. Research into reasoning models is already focused on the problem of apparently logical answers that are nevertheless unsupported by the underlying evidence. [aclanthology.org], [frontiersin.org]

That matters enormously in training for regulated industries.

Picture a multilingual digital safety instructor rather than a prerecorded presenter.

An engineer could interrupt:

"Why do I have to isolate this equipment?"

The instructor might explain, test their understanding and adapt the next question. An employee confused about a conduct rule could ask for another example. A learner in São Paulo and a learner in Shanghai could explore the same approved policy conversationally in their own languages.

That is potentially transformative.

But a beautifully rendered avatar confidently inventing a safety instruction is considerably worse than a boring PowerPoint slide.

So, the competitive advantage may not ultimately be the most photorealistic face. I strongly believe it will be the trusted, multilingual knowledge that powers it, accompanied by strong governance such as ISO:27001 and auditable, human oversight.

Future-proofing principle No. 2: build training content today as if tomorrow it will have to answer questions.

Structure it. Verify it. Maintain it. Control the terminology. Know which version is authoritative.

3. Don't automate the humans out of learning

There is an understandable fascination with making digital humans increasingly indistinguishable from us.

I'm not convinced that should be the ultimate objective.

People may enjoy interacting with AI, but human connection still has qualities technology does not automatically reproduce. My time teaching officers at police college and advising governments on how to improve public safety taught me a valuable lesson, make learning personal, and people will understand, then apply that learning. Introducing a variety of teaching methods like video case studies, quizzes and role plays is crucial – and being there with the learner is paramount.

Need proof? Well, a large 2026 study of more than 50,000 conversations found higher measured social presence in human-to-human exchanges than human-AI exchanges, even though the AI conversations demonstrated greater structural continuity. Research into human-AI collaboration also suggests that social and interpersonal processes remain important even as AI becomes a more capable collaborator. [frontiersin.org] [onlinelibrary.wiley.com]

That points towards what I suspect will become a powerful model for learning:

AI first. Human when it matters.

Imagine an avatar delivering mandatory training at scale in 40 languages. Most learners complete it independently.

But someone has a difficult question.

Or challenges an interpretation.

Or simply wants to talk to someone who understands the subject.

Instead of the machine bluffing its way through, the learner can instantly reach a real expert.

That isn't a failure of automation.

It is good automation.

AI does what it is brilliant at: availability, consistency, personalisation, speed and scale. Humans provide judgement, expertise, reassurance and accountability when they are actually needed.

2033 may be closer than it looks

Seven years ago, I was sitting in a BBC studio talking about a fraud pandemic.

Today Guildhawk can show that conversation again with both participants apparently speaking Mandarin, or in 100 other languages for that matter, thanks to advances in AI.

Seven years from now, I suspect we'll look back at today's avatars in much the same way we now look at the AI of 2019 and that first Synthesia avatar of me when I gave the FAP Chairman’s address at the AGM during lockdown.

The winners won't necessarily be the organisations with the flashiest digital humans.

They will be those that spent the intervening years building trusted knowledge, protected intellectual property, accurate multilingual content and intelligent ways for humans and AI to work together.

For compliance and safety leaders, that work can start now.

Because the future of global training isn't simply translating the same video into more languages.

It is about keeping people and our planet safe by making trusted knowledge available to every employee, in their own language, wherever they are, with a human never too far away.

And that is a future worth building.