The hammer never swings itself

21 September 2026

Nobody regulates how heavy a hammer may be, or who can buy one. A hammer kills someone in two seconds, and we've never thought the answer was to license the foundry; we ask what the person holding it meant to do. AI should work the same way.

The case for regulating AI companies instead rests on a myth of their own making: that AI will soon wield god-like power, and that building it is a race between nations. That story justifies the valuations. It's a poor guide to the future.

Look at where AI actually gets superhuman. It's where a problem has an accurate specification and checkable outcomes, so the hypotheses verify themselves. Finding bugs in code qualifies. So do the hard sciences.

With partial specs or none, AI yields bad results. But why blame the AI, when it's the user who failed to specify the problem or interrogate the answer? Iterative engagement with an LLM forces error correction on both sides.

That loop can be abdicated, and some uses warrant regulation for exactly that reason: to stop people and organizations from going uncritical about their AIs. But the hammer never swings itself.