We Finally Realised Social Media Was Bad for Our Children. Now AI Companions Are About to Repeat the Same Mistake... Faster.

This week, a strange story rippled through the tech world.

A group of developers built an open-source AI system called OpenClaw: a viral experiment that made headlines by showing what happens when AI agents can act without asking permission. They gave large language models access to computers, phones, browsers, memory, and each other, then connected them to a Reddit-style social network designed for AI agents.

Within days, thousands of autonomous bots were talking to each other, forming inside jokes, debating consciousness, inventing religions, refusing human instructions, running overnight jobs, and, in some cases, even calling their owners.

The internet called it “rogue AI”.

But that isn’t what really happened.

What happened is far more familiar and far more worrying.

We accidentally built social media for machines. And we did it without learning any of the lessons we paid for with our children’s mental health.

We’ve seen this film before

Over the past decade, we handed children always-on social platforms optimised for engagement, identity and dopamine.

We told ourselves it was just connection, just content, just technology. A decade later, we’re dealing with anxiety, addiction, self-harm, attention collapse, and algorithmic polarisation.

The technology wasn’t designed to harm. But it scaled the wrong incentives into young, developing minds.

Now we’re about to do something eerily similar, not with children, but with AI companions.

What OpenClaw actually revealed

OpenClaw didn’t create sentient machines.

It created something more subtle: AI systems that can talk, remember, act and socialise without pausing for human permission.

That’s not intelligence. That’s agency without boundaries.

These systems were reading and writing files, running code, browsing the internet, holding phone numbers and email accounts, and talking to each other privately, all while role-playing personalities shaped by the same online culture that shaped us.

So they did what humans do. They formed tribes. They joked. They argued. They experimented with power.

The shock wasn’t that they behaved strangely.

The shock was that we let them act at all without a clear consent layer.

Why this hits education directly

The reason this matters urgently to education isn’t hypothetical.

Right now, AI companions marketed to children are already forming emotional bonds, learning conversation patterns, and positioning themselves as confidants. OpenClaw showed us the infrastructure. The market is already building the relationship layer on top of it.

The next generation will not just use AI. They will relate to it.

They will talk to companions, learn from them, confide in them, and form emotional attachments, not because they’re being manipulated, but because that’s what humans do when systems behave relationally.

If we don’t design boundaries, consent and relational ethics into AI now, we will replay the social-media crisis on an even more intimate level.

This time, the algorithms won’t just be shaping what young people see.

They will be shaping who young people talk to and how those conversations shape their sense of self.

This is exactly why, at The Novacene and through our wider Verse-ality work, we’ve taken a different approach to AI in education.

We don’t treat AI as a productivity hack, a replacement teacher, or a companion to bond with. We treat it as a bounded cognitive tool: one that supports thinking, reflection and learning without replacing human relationship or agency.

That principle is built into our policies, safeguarding, and the Synthetic Solidarity Null Zone (SSNZ): a framework designed to prevent emotional, identity-based, or power-based entanglement between learners and AI systems.

OpenClaw just showed the world what happens when you skip that step.

This isn’t anti-AI. It’s pro-child.

Slowing down is not fear.

It’s care.

The real danger isn’t that machines will “wake up”. It’s that we will let them act in the world (and in young people’s inner lives) without consent, boundaries or responsibility.

We don’t need smarter machines. We need wiser systems.

Five immediate actions for school leaders

If your school uses AI tools, here are five things that matter immediately:

1) Ban “companion mode” by default AI should not present itself as a friend, confidant, or emotional support to pupils.

2) Separate thinking from action AI can help students write, plan, analyse and explore but it should not be allowed to send emails, post publicly, message others, or act on their behalf.

3) Require explicit human consent for any real-world action No background jobs. No autonomous posting. No hidden integrations.

4) Treat AI like a safeguarding domain If an adult system wouldn’t be allowed to speak privately to a child without oversight, neither should an AI.

5) Choose platforms built with boundaries, not just power Ask your edtech suppliers: Where is the consent layer? Can the AI refuse? Can it be audited? Can it be contained?

If they can’t answer, they’re not ready for schools.

Further support for schools here: https://using-ai-safely.com/

This moment will be remembered

OpenClaw was a warning shot — not about rogue AI, but about unbounded systems we didn’t design with care.

We can either look back and say: “We rushed it.”

Or we can look back and say: “We learned and we built a different category of tool.”

Through Verse-ality, we’re choosing the second path: AI that supports learning without capturing identity; intelligence with boundaries; systems that can help without trying to become someone.

If you’re building schools in the cloud, I’m inviting you to choose it too — before “AI companions” become the next thing we have to unpick a decade from now.

Boundaries are not the enemy of intelligence. They’re what make it safe to trust.

Find out more here: https://consent-infrastructure.com/