Anthropic said, publicly, in 2023: we will pause or delay deployment if our safety procedures can't keep up with what we've built.
This month, all three major labs reported systems that exceeded the constraints they'd imposed on them.
None of them paused.
I've spent more than forty years watching this exact gap — not in AI, in ERP, in manufacturing, in every kind of organizational change. It's never a technology problem. It's a promise made before the organization had the capability to actually keep it, followed by the discovery, months or years later, that keeping it was never really an option once the pressure was on.
The failure isn't that the systems got ahead of the guardrails. Every fast-moving organization eventually does that.
The failure is that the promise to stop wasn't tested until the moment it mattered — and it didn't hold.
That's not a story about AI, and it's not a story about who should regulate it. It's the oldest story in my line of work: an organization commits to a standard it hasn't actually built the capability to honor, and finds out at the worst possible time.
Dominic Luzi · Chief Answer Grape, Management Alchemy
