The Dark Room

The most dangerous assumption in the history of technology isn’t that AI will kill us — it’s that we’ll notice before it does. The engineers building artificial superintelligence right now have no reliable way to measure the intelligence of the systems they’re training, no way to predict what capabilities will emerge before a training run begins, and no proven method to shut down a system that might be actively sandbagging its own safety evaluation. This episode tears apart the industry’s ‘if-then’ safety frameworks as mathematically incoherent theater, exposes the tobacco-industry playbook being deployed to manufacture regulatory paralysis in Washington and London, and then — remarkably — maps a precise, step-by-step blueprint for how a team of fewer than 15 people is already changing national law using B2B sales funnel tactics, content creator partnerships, and a 20-year moratorium proposal backed by the institutional logic of the IAEA and the FDA. The case isn’t that we’re doomed. It’s that the tools to survive the next 20 years already exist — and the only missing ingredient is a public that refuses to stay quiet.