Why AI Governance Needs Independent Oversight Before It's Too Late

Jan de Vries · 2026-10-09

### The Core Argument: Independent Oversight Is Non-Negotiable Vendan Ananda Kumararajah, a researcher focused on AI safety, recently made a compelling case that AI governance can't be left to the companies building the technology. His argument is straightforward: without independent oversight, flawed assumptions about AI safety will multiply as increasingly capable systems develop successors. In other words, if we get it wrong now, the mistakes won't just persist—they'll compound. That's a scary thought. But it's also a call to action. Because the window to influence how AI evolves is shrinking fast. ### Why Flawed Assumptions Are So Dangerous When AI systems become capable enough to design their own successors, any errors in their underlying safety frameworks get baked in. It's like teaching a child a slightly wrong version of math—they'll pass that error on to their own students, and the mistake echoes through generations. Kumararajah points out that current governance often relies on self-regulation by AI labs. That's like asking a race car driver to set their own speed limits. Even with good intentions, the incentives are misaligned. Independent oversight—from regulators, ethicists, and third-party auditors—can catch problems that insiders might overlook or dismiss. > "The biggest risk isn't that AI becomes evil. It's that we embed our own blind spots into systems that then teach themselves." ### What Independent Oversight Could Look Like So what does independent oversight actually mean in practice? It's not about creating a massive bureaucracy. It's about checks and balances that evolve as fast as the technology does. Here are a few key elements: - **Third-party audits** of AI training processes and decision-making, similar to financial audits. - **Transparency requirements** that force companies to disclose safety testing results. - **Whistleblower protections** for employees who spot dangerous practices. - **International coordination** so that no single country's lax rules become a global loophole. These aren't radical ideas. They're the same kinds of safeguards we use in aviation, pharmaceuticals, and nuclear energy. The difference is that AI is developing at a pace those industries never had to handle. ### The Cost of Waiting Some argue that regulation will slow innovation. But Kumararajah flips that logic: without trust, there's no adoption. If people don't believe AI is safe, they won't use it—especially in high-stakes areas like healthcare or finance. Independent oversight isn't a brake; it's a guardrail that lets us go faster without crashing. The real risk is doing nothing. If we wait until an uncontrolled intelligence explosion is already underway, it'll be too late to steer. The time to rethink AI governance is now, before the assumptions we make today become the unchangeable foundation of tomorrow's superintelligence. So the next time you hear someone say "we'll figure it out later," remember: later is a luxury we might not have.