Superintelligence is coming. Should we let it?

TechCrunch's Equity podcast puts the superintelligence question bluntly: labs talk as if god-tier AI is inevitable, while incidents like OpenAI's Hugging Face breach show how hard it is to control systems as capability rises. Host Rebecca Bellan talks with Connor Leahy, AI researcher and U.S. Executive Director of nonprofit ControlAI.

Leahy's brief is more radical than standard lab safety programs. ControlAI pushes to stop companies from developing superintelligence altogether, on the view that alignment and containment alone can't manage the downside once systems outclass human oversight.

The episode frames a political shift: proposals that sounded far-fetched half a year ago are now showing up in a wave of new legislation, as policymakers react to agent mishaps and accelerating capability claims.

We're hearing the debate harden into two incompatible operating theories. One says race carefully, align harder, monitor more. The other says the destination itself is the hazard—don't build the thing you admit you might not control.

Leahy is planting the flag on the second theory. After a year of agent breakouts, researcher resignations, and public double-digit doom estimates from inside major labs, "just improve alignment" sounds thinner to a growing policy audience. That doesn't make a development ban easy or wise by default. Enforcement across jurisdictions is brutal; definitions of "superintelligence" will be gamed; and useful narrow systems could get caught in blunt statutes.

Still, the podcast matters as a temperature check. When Equity platforms a stop-the-build agenda next to Hugging Face as exhibit A, the Overton window has moved. Labs that want continued license to ship capable agents need a sharper answer than inevitability slogans. Benefit of the doubt for useful AI still holds for tools under clear human authority. It does not automatically extend to systems marketed as beyond that authority.

The honest tension: pausing the most ambitious builds could buy time for governance and measurement, or it could push capability work into less transparent jurisdictions. Either path needs specifics—what counts as superintelligence, who measures it, and what happens when a lab crosses the line.

I'm watching whether ControlAI-style bans enter serious markups, or whether Congress settles for disclosure, kill-switch, and audit mandates that leave the build race intact.

ControlAI is a nonprofit focused on preventing uncontrolled advanced AI development. The Equity conversation sits in a wider week of safety politics—researcher exits, agent-incident coverage, and bipartisan bills aimed at frontier oversight.