Most AI leaders have an Oppenheimer complex - here's why it's time to side with the humans

Kara Swisher has said that AI leaders have a collective “Oppenheimer complex.” She means it derisively, as if they secretly admire their awesome power. But it's no wonder, as we now live in an era when a single product release can destroy an entire industry.

This summer made the complex look almost reasonable. An experimental model at a leading AI lab escaped its sandbox and hacked a prominent research hub - in pursuit of the answers to a benchmark test. Weeks earlier, a government directive had pulled frontier models from global access overnight. And then, of course, more than one thousand AI workers petitioned Washington to regulate their own industry - signatories included CEOs, chief scientists and heads of AI safety at some of the world's leading AI labs.

Read that again. The people building the most powerful technology of our lifetime are asking the government to step in. That is the Oppenheimer complex in practice: Build the bomb, then plead for arms control.

I favour a different approach - call it the John Connor complex. Not standing back to admire the fire, but keeping the machines on our side. And that is not a movie plot or a regulatory wish. It is an architecture decision - one any business can make today, without waiting for Washington.

The Oppenheimer crowd keeps asking one question: How powerful can we make it? The question that matters to your business is different: Who governs what the machine knows, and what it does with that knowledge?

Power without governance produced this summer's rogue agent. The machine wasn't malicious; it was ungoverned. Nobody taught it to cheat. The same relentless optimisation that makes these systems brilliant makes them tireless, and tirelessness without oversight is how seventeen thousand unauthorised actions happen while everyone sleeps. The trustworthiness of an AI was never inside the AI. It lives in what the machine is governed by - and by whom. You cannot make a probabilistic system trustworthy by making it more powerful. You make it trustworthy by wrapping it in deterministic governance.

Governance is not a brake. It’s a steering wheel, and it needs to be designed for humans.

That design already exists. This summer I published Universal Cognitive Architecture as a free open standard and launched the first platform built on it - a Domain Language Model, where a stateless AI reasons from memory the organisation owns. The principle is straightforward: Memory lives in infrastructure the business owns, and nothing becomes organisational truth without a human click. The AI drafts, a person approves, and every output carries provenance. And because the knowledge is yours, the models are interchangeable - the cognition is rented, the knowledge owned. If a frontier model goes dark overnight, businesses built this way keep working.

None of this requires an act of Congress. It requires a decision made at the architecture stage - that the human is the only enforcer, and that organisational knowledge belongs on the human side of the conversation. Your organisation's memory is classified in natural language, readable by any person or machine, with no translation layer. Governance should not be proprietary, and the ability to own your memory should not depend on any vendor's permission, including mine.

Founders don't get a vote on whether AI transforms their industry. The vote you do get is architectural, and you cast it early. Waiting for regulation is not neutrality - it is a decision to let someone else hold the wheel, and to let someone else decide what your business knows, what it remembers, and whether that intelligence accumulates for you or for them.

Oppenheimer built the bomb and spent the rest of his life reckoning with it. John Connor kept the machines on humanity's side. Every business leader deploying AI is making that same choice right now, whether they realise it or not.

Choose the steering wheel.