AI Governance at a Crossroads: Why Independent Oversight Can't Wait

Jan de Vries · 2026-10-09

Vendan Ananda Kumararajah isn't mincing words. He's arguing that if we don't rethink how we govern AI, we're staring down the barrel of an uncontrolled intelligence explosion. That's a heavy claim. But the more you sit with it, the more it feels less like sci-fi and more like common sense. Here's the core of his argument: AI systems are getting more capable. That's not news. What is news is how those systems are starting to develop successors. And when they do, any flawed assumptions baked into their governance structures don't just stick around. They multiply. Like a virus in a crowded room. ### Why Independent Oversight Isn't Optional Think about how most companies handle AI safety today. It's often internal. Self-policed. Which is a bit like asking a student to grade their own exam. They might be honest. They might not. But you'd never bet your future on it. Kumararajah's point is simple but sharp: we need independent oversight. Not because companies are evil. But because the stakes are too high for good intentions alone. > "Flawed assumptions don't stay contained. They replicate. And when they're embedded in systems that build other systems, the replication happens faster than we can correct course." That's the part that keeps me up at night. Not the AI itself. The assumptions we feed it. ### The Successor Problem Here's where it gets really interesting. We're not just building AI that does tasks. We're building AI that builds better AI. That's the successor problem. Each generation learns from the last. But if the last generation had a blind spot, the next one inherits it. And the one after that inherits it too. Only now it's harder to spot. You've seen this in companies, right? A bad process gets passed down. Nobody questions it because "that's how we've always done it." Now imagine that at machine speed. - Flawed assumption enters the system - System trains a successor - Successor inherits the flaw and builds on it - Flaw becomes invisible infrastructure That's not a hypothetical. That's the default outcome without independent checks. ### What Good Governance Actually Looks Like Independent oversight doesn't mean slowing down innovation. It means building guardrails that don't depend on the same people who benefit from ignoring them. We need third-party audits. We need transparent reporting. We need governance that evolves as fast as the tech does. That's not bureaucracy. That's survival. Kumararajah isn't calling for a halt. He's calling for a rethink. And honestly? He's right. The question isn't whether AI will shape our future. It's whether we'll shape how it does. If we don't, we're not just risking an intelligence explosion. We're risking one we can't control. And that's a gamble none of us should be willing to take.