Nobody Is Going to Slow Down

Recently, Anthropic’s CEO Dario Amodei published a manifesto on his personal domain arguing that the industry should pace frontier AI development. He laid out a three-step plan for responsible pacing. It was thoughtful. It was measured.

Sam Altman (OpenAI/ChatGPT) and Elon Musk (xAI) both publicly agreed that this was the right idea. The LA Times covered it as a rare moment of alignment across rival AI labs. The internet nodded along. People wrote threads about it. Everyone seemed to agree that something important had just been said.

OpenAI even delayed their 2026 IPO the same day, which certainly doesn’t contradict the narrative.

I think none of it matters.

It Is Publicity Theater

If Anthropic were serious about pacing frontier development, it wouldn’t be announced on the CEO’s personal domain. It would be a press release from the legal team and the board. It would come with binding commitments, not personal opinions. (OpenAI delayed their own IPO on the same day, which is certainly an interesting coincidence.)

A personal blog post is a pressure valve. It lets the CEO signal thoughtfulness without committing to anything the company has to actually do. It’s algorithm massaging. It’s IPO theater.

I wrote about focusing on the solution rather than the hype earlier this month. The same principle applies here. Real commitment looks different. It has skin in the game. This doesn’t.

Companies Won’t Slow Down Voluntarily

Companies in 2026 don’t voluntarily slow down unless forced to by the government. And we all know this government isn’t going to do a damned thing to slow anyone down.

The regulatory environment around AI in the US is basically a suggestion at this point. There are voluntary frameworks. There are executive orders. There are lots of meetings. But there is no real enforcement mechanism. No teeth. No consequences.

Even if every CEO in the room wanted to slow down tomorrow, the competitive landscape makes that impossible.

The Prisoner’s Dilemma Is Real

None of these companies are going to slow anything down because the companies with increasingly competitive capabilities aren’t going to slow down either.

This is a classic prisoner’s dilemma. If Anthropic slows down and OpenAI doesn’t, Anthropic loses. If OpenAI slows down and Anthropic doesn’t, OpenAI loses. If both slow down, they’re fine. But trust doesn’t exist in this space. The incentives are structured so that not slowing down always wins for the individual actor. And that’s completely ignoing the increasing pressure coming from companies like Alibaba, Meta, Deepseek, and a growing list of others.

Virtually every company formed since mid-2025 has “AI” in their name somewhere. The rest have it in their core mission. Nobody’s slowing anything down. The momentum is just too big. We’re strapped to this rocket.

Where I Stand

Is this the right thing to do? Am I qualified to answer that question? No.

I don’t live out in the thin air of what giant LLMs can do when given unlimited compute, unlimited data, and unlimited ambition. I live down next to the road, helping people make the small, important work a little easier in their day-to-day. I see the friction. I see the value. I see both.

I do know bullshit when I read it. And this whole exchange smells like a cattle pasture.

What I Actually See

Here is my read on what is happening:

• Anthropic is stoking the publicity fires. IPO season doesn’t care about your safety framework. It cares about your narrative.
• OpenAI is willfully or ignorantly colluding. Whether they believe in the pacing idea or just need to stay in the conversation, they’re contributing to a theater production.
• Elon Musk is just prattling on. He has been doing this for fifteen years. The pattern is recognizable.
• Nobody is being serious about being serious about what may be a very real problem.

There is a difference between talking about responsibility and actually practicing it. The people who are practicing it don’t need a manifesto on a personal blog. They’re too busy building.

The Nuance

I’m not saying AI safety doesn’t matter. I’m not saying the concerns are overblown. I’ve written about the gap between AI hype and real value before. This feels different though. This isn’t about hype in the industry. This is about hype as a communication strategy.

The real work of AI safety and responsible development happens in engineering decisions, in system design, in the mundane details of how models are trained and deployed. Not in CEO-to-CEO agreement on personal blog posts.

The people doing the real work in this space know the difference between theater and substance. They’ve been building in the trenches while the headlines got their five minutes of glory.

So What?

The momentum in AI isn’t slowing down. It probably won’t slow down for a long time. If you’re working in this industry, the pragmatic move isn’t to wait for a pause that isn’t coming.

It’s to focus on the work that matters. The work that helps actual people. The work that builds systems that are honest, reliable, and useful. The kind of work that doesn’t need a manifesto to justify it because the value is obvious to anyone who uses it.

That’s the kind of seriousness I respect. The kind that shows up in shipping software, not in press releases. Am I being pessimistic here? Yes. Do we live in late-stage capitalism? Again, yes.