Do you trust the data your AI acts on?

As AI moves closer to the edge, organisations are investing heavily in protecting AI models. But we securing the right thing?

Artificial Intelligence (AI) is transforming the way we design, operate and protect connected systems. Whether it’s autonomous vehicles, intelligent surveillance, industrial automation or defence platforms, AI is increasingly making decisions that were once reserved for humans.

Naturally, much of the conversation has focused on the AI model itself. How do we make it more accurate? Faster? More efficient? How do we prevent it from being stolen or manipulated?

These are important questions.

But perhaps they’re not the first questions we should be asking.

Before an AI model can make a decision, it must first receive information from the outside world. Cameras, sensors, radar systems and connected devices continuously feed data into the model, which then interprets that information and decides how to respond.

This raises a more fundamental question.

Can you trust the data your AI acts on?

Because if the data entering an AI model has been manipulated, spoofed or replaced, even the most advanced algorithm will produce the wrong outcome. The AI isn’t failing. It’s simply making decisions based on information it believes to be true.

For years, cybersecurity has concentrated on protecting software, networks and, more recently, AI models. Yet the integrity of the data itself often receives far less attention.

As AI becomes increasingly embedded in mission-critical applications, this assumption deserves to be challenged.

The future of trustworthy AI may depend less on building smarter models and more on establishing trust long before AI begins its analysis.