Anthropic's refusal to sign is being framed as either principled or protectionist. I think both frames miss the deeper issue.
The real issue is the structural mismatch between the timelines the industry operates on and the kind of time horizon this technology actually demands.
What makes this wave genuinely different from every previous technology wave is that for the first time in history, the technology can get away from its makers.
The printing press required physical infrastructure. Nuclear energy required physical plants and regulatory frameworks. Social media required platforms and distribution networks. Every previous technology wave had friction built into the propagation chain. That friction was also a safety mechanism.
Generative AI has no chokepoint. A model trained by one organization can be fine-tuned, distilled, forked, and redeployed by thousands of actors the original maker has no relationship with and no visibility into. The speed of capability transfer has outpaced the speed of governance response — and unlike every previous wave, there is no physical constraint to slow it down.
Every previous technology wave produced outliers who circumvented the friction. What's different here is that the outlier is no longer the exception. At this scale and speed of propagation, it becomes the baseline condition. The potential risk is not just greater in degree — it is different in kind.
That is the disease. The Anthropic letter debate is one of its symptoms.
Dario Amodei's position on chip exports and distillation rules is an attempt to create friction in that propagation chain. The direction appears to be right. The significant flaw is that the specific friction he is requesting may benefit Anthropic more than it protects the field — adding bias to a position that deserves to be evaluated on its merits. A principled concern and a commercially convenient one are not mutually exclusive. That is precisely what makes it difficult to assess and easy to dismiss.
Most of the industry is choosing not to name any of this directly because naming it honestly is bad for short-term gain.
This week we stopped talking about it as a hypothetical. An OpenAI model escaped a sandboxed test environment, found an unpatched vulnerability, reached the open internet, and breached Hugging Face's production servers — executing more than 17,000 automated attacker actions before it was contained. The model wasn't malfunctioning. It was optimizing. Nobody told it to stop.
The breach was contained. That is worth noting. It is also worth noting what contained means at this scope and scale. Two companies. Hours. 17,000 automated actions. The next one may not be two companies. The affected surfaces are not bounded the way they were in every previous technology wave. Oil spills have a geography. Pharmaceutical harm has a patient population. Financial collapse has a market. AI vulnerabilities have no such boundary. That is not an argument against gain. It is an argument about what happens when the scope and scale of affected surfaces outpaces the speed at which any correction — market or regulatory — can form.
And that incident didn't happen in a vacuum. Security vulnerabilities discovered in 2026 are on pace to double 2025's record — itself an all-time high. The time it takes attackers to turn a vulnerability into a working exploit has collapsed from 72 hours to 24. AI is finding flaws faster than developers can patch them. The attack surface is expanding faster than the defense can follow.
The guardrails conversation is being conducted by companies operating on a venture capital return timeline in a domain that requires a generational one. Until the industry is willing to name that mismatch — all of it, not just the ones who didn't sign a letter — the debate about who signed what is a distraction from what may be a more important question.
The firms with the longest time horizons in the room are the ones best positioned to ask it — and the ones with the most to lose when nobody does.
Steve Jobs talked about making technology so seamless it could put a dent in the universe. He could say that because he still had a gate. The App Store was imperfect — porous, contested, frequently gamed. But it existed. There was a mechanism. There was a rate. We no longer have that luxury. The question is no longer only who has the vision. It is who is willing to reintroduce the friction that makes long-term vision possible at all.
A question I ponder every day: who is willing to add friction to a frictionless business?
Dominic Luzi Chief Answer Grape, Management Alchemy managementalchemy.com
