newseditor 
By Andreas Hassellöf, CEO, Ombori & Phygrid
We are living in an era where the boundaries of reality and fiction are increasingly blurred by AI. Just last year, a fake AI-generated image of an explosion near the Pentagon briefly rattled financial markets. Apple had to disable its new AI news feature after it started hallucinating headlines. These incidents didn’t just make headlines, they exposed something deeper – the widening trust gap surrounding AI.
If businesses are going to benefit from this technology, we have to confront this trust issue head on. We need to design systems that are secure, predictable, and explainable. That means being deliberate about how AI is used and where it runs. As deepfakes and misinformation continue to evolve, so must our defenses through a mix of policy, tech choices, and architecture.
The Real Threats Are Already Here
Deepfake attacks are no longer niche. We are seeing voice and video forgeries used to impersonate executives, approve fake wire transfers, and hijack conversations. One major European firm lost over €200,000 after a CEO was spoofed using synthetic audio. In the Middle East, hackers recently aired AI-generated misinformation on live television to influence public sentiment.
But fraud is only the tip of the iceberg. AI models can now be manipulated subtly, without anyone breaking into them directly. This is where data poisoning comes into play. Bad actors are learning that it is easier to get AI to speak negatively about a competitor than to promote their own brand. Public data sources get seeded with biased content, and over time, the AI unknowingly starts to internalize it. This is happening today. It is an arms race of narrative shaping, and most organizations aren’t prepared.
If You Don’t Control the Stack, You Don’t Control the Outcome
A lot of enterprises focus on prompt policies and user guardrails, but that’s only part of the solution. The more important question is: do you actually control the infrastructure where your AI is running? If the answer is no, you are flying blind.
This is a big part of why we built Phygrid, which is now an Ombori company. Phygrid takes a very different approach from the traditional cloud-heavy models. It brings AI to the edge close to where people actually interact with technology. We designed it so companies can deploy models directly on devices in stores, venues, airports, and hospitals. The result is faster response times, improved reliability, and most importantly, tighter control over data and model behavior.
Why the Edge Matters More Than Ever
Cloud has been the default for a long time, but it’s not always the right answer. When every interaction has to bounce between a sensor, a cloud data center, and a local display, you introduce latency, increase risk, and complicate compliance.
With Phygrid, AI models run on-site. If a store loses internet, the AI still works. If the network lags, customers still get real-time answers. It’s not just about speed, it’s about control. No external provider is logging your biometric feeds or repurposing your interaction data. You own it all.
This model is also more resilient. You’re not depending on a third-party uptime guarantee to keep your in-store systems functional. And when it’s time to update models or deploy fixes, you can do it centrally, instantly, and with confidence. That’s a huge shift in how physical AI is managed.
Moving From Mistrust to Momentum
There is a real appetite for AI in the physical world. Cities want smarter infrastructure, retailers want more engaging customer experiences, and healthcare providers want to automate without risking privacy. But there is hesitation, and rightly so. The stories of chatbots leaking data or AI assistants inventing headlines are still fresh in everyone’s mind.
We believe the way forward is to anchor AI in a framework that prioritizes sovereignty, transparency, and security. That includes choosing platforms that let you process data on the edge instead of sending everything to the cloud, being intentional about the data you use to train and fine-tune models, and being ready to detect when your systems are drifting off course.
What Responsible AI Looks Like in Practice
Here are some principles that guide our thinking:
- Control where AI runs. Whether it’s an airport kiosk or a smart shelf in a store, the closer the model is to the action, the better the user experience and the lower the risk.
- Don’t send everything to the cloud. Only move the data you actually need. Local processing dramatically reduces exposure and regulatory headaches.
- Expect adversaries. Build systems assuming someone will try to manipulate them—either through poisoned data, misleading feedback loops, or model exploits.
- Design for transparency. If your customer doesn’t know they are talking to AI, that’s a failure. If your team doesn’t know what the model is doing, that’s a crisis waiting to happen.
- Invest in resilience. Networks fail. APIs change. Policies evolve. Make sure your AI keeps working no matter what.
Why This Matters Now
There’s never been more money flowing into AI. Enterprises and governments are investing aggressively. But trust is not scaling at the same rate. That’s the gap we need to close.
If we get this right, AI will change how we interact with the physical world in extraordinary ways. If we don’t, the fallout will be just as dramatic.
At Phygrid, we’re betting on a future where edge-native AI isn’t a niche architecture—it’s the new normal. Where deploying smart, secure systems doesn’t mean sacrificing performance or privacy. And where brands feel in control, not just of the user experience, but of the entire AI lifecycle.
That’s how we build trust. That’s how we move forward.
