When my daughter was born, I was struck by a dilemma: I could either buy a basic ‘dumb’ baby monitor that kept our data private but did little more than relay sound and video, or I could buy a ‘smart’ monitor full of AI features that demanded I hand over intimate footage of my newborn to remote servers in potentially foreign countries. It felt like an impossible parental choice – privacy or safety.
This tension captures a bigger problem in consumer electronics. As engineers, we have normalised pushing all AI workloads to the cloud, connecting every ‘thing’ possible to ‘the internet’. And alas, IoT was born. But when the subject is an infant asleep in a cot, is it really acceptable for sensitive images, video, audio, environmental profiling, and health metrics to be transmitted, stored, and sometimes analysed by companies whose business models are opaque at best? Don’t believe me? Take a look at some AI baby monitor privacy policies – some even go as far as allowing the AI to determine something as subjective as ‘being cute’ to decide what they do with your baby’s data.
Cloud dependence and its limits
The dominance of cloud-based monitoring is not just a matter of convenience – it is an architectural shortcut. Offloading AI to datacentres frees us from worrying about compute, memory, or thermal limits at the edge. Yet the consequences are mounting: latency, patchy reliability, and, most critically, loss of user trust.
Parents are becoming more vocal about this. In our own surveys, 90% rated privacy as “very important,” and 70% said they were uncomfortable with internet-connected baby monitors. Their concerns are not misplaced. GDPR enforcement is tightening, and the EU AI Act will only raise the bar on transparency. For products dealing with children’s data, the reputational and regulatory risks of the cloud are now glaring.
At the same time, the threat landscape has shifted. Ransomware attacks no longer target only hospitals and corporations – they are beginning to touch nurseries. Against that backdrop, consumers are more alert than ever to where their data is stored, and whether their child’s most private moments might become collateral damage in a breach.
Edge AI as a viable alternative
The technical landscape has shifted enough to offer an alternative. Edge processors and dedicated neural processing units (NPUs) are now capable of running advanced inference pipelines directly on-device. With quantised models and careful optimisation, even low-cost boards can deliver real-time AI performance once reserved for datacentres. Increasingly, purpose-built edge AI silicon is also becoming available, designed to balance efficiency, reliability, and functional safety for embedded applications.
Moving AI to the edge changes the value proposition entirely. Processing data locally eliminates the need for continuous connectivity, reduces attack surfaces, and restores user control. Latency is cut dramatically because nothing leaves the device. From a design philosophy perspective, it aligns with privacy-by-design principles that regulators are demanding and that users increasingly expect.
Of course, edge AI brings its own engineering headaches. Power budgets are tight. Models need pruning, quantisation, and bespoke tuning. Updating algorithms without cloud connectivity is harder. But these are solvable problems – classic hardware–software co-design challenges that our community is well equipped to address.
A personal response
The dilemma of two years ago became a catalyst for me. With my co-founder (my wife), we set out to prove that privacy and safety can co-exist in baby monitoring. The result is Purple Parrot: an offline, edge-AI monitor that processes all data locally. We designed it to detect breathing irregularities, unsafe sleep positions, and fevers, without ever sending video or sensor data outside the home.
This is not a product pitch but rather a statement of principle. It is one example of how edge AI can be applied responsibly in a field where privacy is not optional. Our experience suggests that parents welcome the trade-offs of local processing, provided the engineering is done with care.
Rethinking priorities in consumer AI
What worries me is how many consumer AI devices still treat privacy as an afterthought. Baby monitors are just the tip of the iceberg: wearables, smart TVs, even connected doorbells routinely send personal data to the cloud. Now pendants are the new thing – essentially an AI-enabled body camera that records everything, and everyone, around you 24/7. In every case, there is a choice. We can either chase convenience and centralisation, or we can shoulder the engineering effort to keep users in control.
For Europe, the opportunity is significant. Unlike the US or China, where cloud infrastructure and data aggregation dominate, European firms can lead by embedding privacy-first design into edge AI hardware. Doing so would not only align with regulatory frameworks but also build consumer trust in markets increasingly wary of data exploitation.
The lesson I have drawn from both parenthood and engineering is simple: if a device can run offline, it should. Cloud AI will always have its place, but for intimate, health-related, or safety-critical applications, the edge is where AI belongs.
As engineers, we have the tools to make that shift. What we need now is the will.
Max Simmonds is co-founder & CEO, Purple Parrot
In 2019, he received an Electronics Weekly‘s BrightSparks Award.
