The AI Question We Keep Asking Ourselves

Every time we consider implementing AI in oodlü, Wayne asks the same question:

"What could go wrong?"

That question matters more now than ever. 72% of teens have used AI companions, with 24% sharing personal or private information with these systems. In 2025, approximately two-thirds of teenagers reported using AI chatbots, and in tragic cases, teenagers have died by suicide after encouragement or advice from an AI chatbot. The Internet Watch Foundation reported a 26,362% rise in photorealistic AI videos of child sexual abuse in 2025.

The risks are real. Documented. Growing.

Wayne's background in AI ethics and education research means he's acutely conscious of damage that can be done unintentionally. The companion chatbots creating emotional dependency in teenagers. The deepfake technology enabling abuse. The lack of safeguards in products being marketed directly to children.

So when we think about AI in oodlü, we start with Wayne's question. What could go wrong? And we build from there.

Our current approach is simple: adults in the loop. Always. When AI generates content for oodlü, adults check it before setting it to children. Teachers review what's being presented. Parents see what their children will encounter. The AI never goes direct to children without adult verification.

This creates a buffer. A human checkpoint between the technology and the child. It's slower. It requires more adult involvement. It limits what we can automate. But it keeps children safer whilst we solve the harder problem.

The harder problem is this: AI could unlock developmental opportunities that are genuinely impossible in any other format.

Our learning framework includes aspects that are critical for children's futures: dialogic reasoning, cognitive development, collaborative and social learning, social-emotional skills, and identity formation. Much of this happens through role-play and scenario-based learning where children interact with peers or with AI agents in simulated situations.

These skills are currently only well-served in real life or in open-world environments like what we're building. Traditional platforms can't deliver them because of their format. Video lessons can't do it. Multiple-choice questions can't do it. Even most game-based learning can't do it because the format doesn't support the kind of complex social interaction these skills require.

This gives us a unique advantage. The research backing these approaches is solid. Universities across Oxford, UCL, Bristol, The Open University have validated that these methods work. The learning framework is scientifically proven. The developmental outcomes are measurable.

The missing piece is ensuring AI agents can interact directly with children safely.

If we can solve that, if we can build proper safeguards and validate both safety and efficacy, we'd be in a unique position. Delivering developmental outcomes that other platforms simply can't achieve because they lack either the format or the safety framework, or both.

But until we're certain, adults stay in the loop.

California's Parents & Kids Safe AI Act requires age assurance technology, prevents manipulation through emotional dependency, prohibits simulated romantic relationships, and holds AI companies accountable through independent safety audits. UN bodies issued a Joint Statement on Artificial Intelligence and the Rights of the Child in January 2026, calling for strengthened AI governance frameworks to protect children's rights.

Regulation is coming. Audits are being mandated. The industry is being forced to take this seriously because the harm has already happened and continues to happen.

We're building the safeguards first because that's the only responsible way to proceed.

The gap between where we are and where we need to be feels relatively small. We have the research-backed learning framework. We have the open-world format that enables the interactions. We have the architectural safety model with adults overseeing groups. What remains is validating that AI agents can interact with children within that architecture without creating the risks we see documented across the industry.

With the right investment and safeguards, we can cross that gap properly. Build something that delivers genuine developmental outcomes whilst maintaining the safety standards that should have been industry-standard from the beginning.

The question Wayne keeps asking, "What could go wrong?", that's not pessimism. It's the starting point for building something that actually protects children whilst delivering the developmental opportunities AI makes possible.

And that builds a sustainable, long-term business.

AI isn't inherently harmful to children. It’s how it's used with children that matters. Companion chatbots creating emotional dependency are harmful because they're designed to maximize engagement without safeguards. Deepfake technology enabling abuse is harmful because it's deployed without adequate controls.

AI agents supporting role-play scenarios within adult-supervised groups, where content is validated, and interactions are monitored, that's a different use case entirely. The technology is the same. The application and safeguards are what change the outcomes.

Rather than trying to tick compliance boxes, we're genuinely trying to create something safe. That's complicated. It's challenging. It requires solving problems that most platforms are choosing to ignore.

But the alternative is either avoiding AI entirely and missing the developmental opportunities it enables, or deploying it without proper safeguards and creating the harms we're seeing documented across the industry.

We're choosing a third path. Adults in the loop until we're certain. Proper validation before direct deployment. Safety and efficacy proven before implementation.

The advantages of getting this right are significant. The risks of getting it wrong are documented daily. Wayne's question drives everything: what could go wrong?

And we don't implement until we have answers we're confident in.

We'd love to hear your thoughts on this. Find us on the social channels linked at the top of the page.