AI in GNSS: Why Cybersecurity Can’t Be an Afterthought

AI is rapidly becoming embedded in GNSS systems — from signal processing and anti-jam/anti-spoof detection through to autonomous positioning corrections. For most teams, the conversation about AI has so far focused on accuracy, data quality, and regulatory compliance.

That’s an incomplete picture.

AI is not just a data problem or a compliance problem. It is a security problem — and for GNSS-dependent navigation, timing, and infrastructure systems, getting this wrong has consequences that go well beyond a failed audit.

Here’s what every GNSS engineering team needs to understand.

1. Cybersecurity Is a Precondition, Not an Add-On

The Guidelines for Secure AI System Development (2023) state it plainly:

“Cybersecurity is a necessary precondition for the safety, resilience, privacy, fairness, efficacy and reliability of AI systems.”

This is easy to read past, but it’s worth sitting with. It doesn’t say cybersecurity is one of several important properties. It says the other properties — safety, fairness, reliability — depend on it. An AI system that isn’t secure cannot meaningfully be called safe or reliable, no matter how good its outputs look in testing.

2. What “High-Risk” Actually Requires

For systems classified as high-risk — and GNSS-dependent navigation, timing, and critical infrastructure squarely fall into this category — the bar is set across several dimensions simultaneously:

• Accuracy — the model performs to the standard the application demands
• Robustness — performance holds up under degraded, noisy, or adversarial conditions
• Cybersecurity — the model and its data pipeline resist manipulation
• Risk management — failure modes are identified and mitigated, not just documented
• Data governance — training and operational data is controlled, traceable, and clean
• Human oversight — a person remains meaningfully in the loop

3. The Threat Surface Has Moved Inside the Model

This is the shift that catches many teams out.

Historically, securing a GNSS system meant securing the hardware, the RF front end, and the network around it. With AI in the loop, the model itself becomes part of the attack surface.

Consider an AI system trained to detect spoofing or jamming on a GNSS signal. If that model can be subtly poisoned, manipulated, or degraded, it doesn’t just stop working — it can continue to appear to work while quietly failing. A compromised model could mask the very interference it was built to flag, giving operators false confidence at exactly the moment accurate information matters most.

This mirrors a problem GNSS engineers already know well from spoofing: the most dangerous failures are the ones that don’t trigger an alarm.

4. Build Security In From the Start

Retrofitting security after an AI system is deployed is significantly harder — and significantly riskier — than designing it in from the outset. In practice, this means:

• Treating model integrity with the same rigour as data quality and algorithmic fairness — it’s not a lesser concern, it’s part of the same assurance picture
• Defining clear acceptance criteria for AI outputs, so that every decision in the loop has a documented threshold for what counts as a valid, trustworthy result — not just “the model said so”
• Involving security expertise during system design, not only during late-stage testing
• Building incident response plans that account for AI-specific failure modes, including subtle manipulation that may not trigger traditional alerts

5. Securing the Intelligence Layer

As GNSS systems become smarter, more adaptive, and more autonomous, security has to evolve with them. Securing the signal is no longer enough — securing the intelligence layer (the models, the data pipelines, the decision logic sitting on top of the signal) is now just as critical.

For organisations operating in this space, the takeaway is simple: bring cybersecurity thinking into AI design early, treat AI security as inseparable from AI safety, and recognise that the next generation of GNSS resilience depends as much on protecting the algorithms as it does on protecting the antennas.

The Bottom Line

AI brings real capability to GNSS — better spoofing detection, smarter signal processing, more autonomous correction. But every one of those capabilities depends on the AI component itself being secure, monitored, and held to clear acceptance criteria.

If you’re integrating AI into a GNSS-dependent system and want a second opinion on your security and assurance approach, we’re happy to take a look.

Linfinity GNSS is a Cambridge-based team of precision positioning engineers with 20+ years of hands-on GNSS integration and testing experience. We work with organisations across maritime, defence, automotive, and autonomous systems.