Navigating the AI jungle: What responsible, secure AI really means for European businesses.

AI is becoming a bigger part of how businesses operate, particularly in customer service. It's helping organisations automate routine enquiries, support agents and make better use of customer data.

But as AI adoption grows, so do the questions around security, privacy and trust.

Where does your data actually go? Who can access it? And how can you be sure your customers' information isn't being used to train someone else's AI model?

In a recent episode of Puzzel Talks, Frieza Brooks, Senior Product Marketing Manager at Puzzel, sat down with Steven Balding, our AI Engineering Manager, to discuss what responsible AI really looks like in practice, why European data sovereignty matters, and how businesses can navigate an increasingly complex AI environment.

Here are some of the key takeaways from their conversation.

EU data sovereignty: It's about more than where your data lives

When businesses talk about keeping their data in Europe, they're often referring to data residency, meaning the physical location where information is stored.

But as Steven explains, there's more to consider. A business might store its data in a European data centre, but what if the company operating it is headquartered outside Europe? Could another country's laws affect who can access that information?

This is where the distinction between data residency and data sovereignty matters.

• Data residency refers to where data is physically stored.
• Data sovereignty concerns the laws and controls governing that data, including who may be able to access it.

For European businesses handling sensitive customer information, understanding both is essential. But knowing where data lives is only part of the responsibility. How it's used and protected matters just as much.

As Steven puts it: "Just because something can be done does not always mean it should be."

And that's where responsible AI comes in.

What does responsible AI actually look like in practice?

Responsible AI can sound like a broad concept, but much of it comes down to how systems are designed, managed and monitored.

For Steven, it starts with familiar security principles, including anonymisation and encryption. But AI introduces additional challenges.

Here are three areas businesses should pay attention to.

1. Protecting identities goes beyond removing names

Anonymising personal information is important, but removing names and phone numbers isn't always enough. Even when someone's name is removed, seemingly harmless details might still reveal their identity.

Think back to Steven's shop example. If someone visits the same place at the same time every day, that pattern might identify them, even without their name. That's why businesses must consider both obvious personal identifiers and patterns that could expose someone's identity.

2. AI needs safeguards against hallucinations and drift

Generative AI is powerful, but it isn't always predictable. Large language models can sometimes produce information that's incorrect or entirely made up. This is known as an AI hallucination.

In customer service, even a small mistake can have consequences. Imagine a virtual agent giving someone incorrect information about their insurance policy or providing an inaccurate answer to a financial question.

Then there's AI drift, which Steven describes as an AI system moving away from the task or behaviour it was originally intended to perform. For example, if an AI solution is designed to summarise customer conversations, businesses need to make sure it continues to perform that task accurately and within its intended boundaries.

This is where AI guardrails come in. Guardrails are safeguards designed to keep AI operating within defined limits. They can help organisations detect problematic outputs, reduce hallucinations and identify when a system starts behaving unexpectedly.

At Puzzel, Steven and his team are continuously investing in safeguards across our AI solutions, including hallucination detection and drift monitoring.

3. Responsible AI requires continuous monitoring

One of Steven's strongest messages is that responsible AI is never a one-time exercise.

As AI technology develops, new capabilities and potential risks emerge. That means safeguards need to be regularly reviewed and improved. Good AI governance should include clear responsibilities, ongoing monitoring and processes for addressing issues when they arise.

For contact centres, that also means making sure AI continues to deliver accurate, relevant and appropriate responses as customer needs and business information change. Ultimately, trustworthy AI depends just as much on how it's managed as on the technology itself.

Want to dive deeper? Explore what responsible AI really means in customer service, and the practical steps your organisation can take to use AI safely, transparently and responsibly.

How do you know your data isn't training someone else's AI?

This is perhaps one of the most important questions businesses should ask their AI providers. When organisations use AI to analyse customer conversations, generate summaries or automate enquiries, they need confidence that their information is being used appropriately.

So how can they be sure their data isn't quietly becoming part of another company's AI model? Steven explains that Puzzel takes a clear approach to this: “We don't use customer data for cross-customer, generalised model training.” In other words, Puzzel doesn't combine customer information into one shared dataset to train a model used across all customers.

Steven also discusses an ongoing initiative focused on strengthening customer data isolation. The principle is straightforward: data belonging to one customer should remain separate from another customer's data. By strengthening those boundaries, businesses can benefit from AI innovation without unnecessarily exposing their information across different customer environments.

And this reflects a wider principle that Steven believes should guide the industry:

"Ethics and security are not a USP, they're the absolute ground truth that needs to be there."

In other words, protecting customer data shouldn't be treated as a premium feature or competitive advantage. It should be a fundamental expectation of any AI provider.

Five questions to ask your AI provider

If your organisation is investing in AI-powered customer service, these are useful questions to bring into your next conversation with a potential provider:

1. Where is our data stored, and who can access it? Understand both the physical location and the legal and operational controls around your information.
2. Is our customer data used to train shared AI models? Ask how information is processed, whether it's used for model training and how customer data is separated.
3. What safeguards are in place to prevent inaccurate AI responses? Look for clear approaches to hallucination detection, monitoring and controlling AI behaviour.
4. How do you protect personal information? Ask about anonymisation, encryption and how the provider addresses the risk of identifying individuals through data patterns.
5. How are you preparing for changes in European AI technology and regulation? A good partner should have a clear approach to ongoing improvement, security and evolving requirements.

These questions can help you look beyond AI features and assess whether a provider is equipped to handle your customers' information responsibly.

How Puzzel puts responsible AI into practice

Asking the right questions is one thing. Getting clear answers is another.

At Puzzel, we believe businesses should know exactly how their customer data is handled and what safeguards are in place. Here's how we approach the five questions above.

1. Keeping your data secure and in Europe
Your data is stored in EU-based storage, with encryption and redundancy built in. Access is restricted to specifically authorised personnel to help keep your information secure.

2. Keeping your customer data yours
We don't use your customer data to train shared AI models. We only train customer-specific models with your knowledge and consent, and your data is never shared with other Puzzel customers.

3. Building safeguards into AI
We use hallucination detection and active monitoring to identify potential issues and improve AI reliability. We're also continuously investing in stronger safeguards, because responsible AI requires ongoing work.

4. Protecting personal information
We use AI-powered anonymisation and rule-based techniques to remove names, keywords and other identifying information. PII detection is built into our AI features, and we don't link voice data to individuals for identification or tracking.

5. Investing in Europe's AI future
We're investing in European technology infrastructure, including EU-based storage, compute and GPUs. As a European company, we believe in supporting a strong, secure European AI ecosystem.

Building AI that businesses and customers can trust

AI presents exciting opportunities for customer service. But making the most of them means understanding where customer data lives, how it's used and what safeguards are in place.

At Puzzel, we're actively exploring European AI technologies, working to keep customer data hosted in the EU or UK where possible, and building relationships with emerging European providers.

As Steven's perspective makes clear, responsible AI isn't a one-time exercise. It requires continuous investment, oversight and a commitment to protecting the people behind the data.

Want to hear more?

Listen to the full episode of Puzzel Talks, "Navigating the AI jungle: finding your way to responsible, secure AI", where Steven Balding and Frieza Brooks explore EU data sovereignty, AI safeguards and the future of responsible AI in Europe.