TL;DR
The short version
On 12 September 2026, Anthropic CEO Dario Amodei published a ~3,800-word essay, "We Must Pace the Frontier," arguing the industry should deliberately slow capability gains so alignment and safety work can catch up.
The argument was old. What was new: Sam Altman, Elon Musk, Demis Hassabis, and Satya Nadella agreed within hours (The AI Daily Brief). For the first time, the frontier labs signaled they won't decide AI's speed by themselves — and a concrete three-step plan gave everyone something specific to fight over.
Built on the 14 Sep 2026 episode of The AI Daily Brief by Nathaniel Whittemore, plus Amodei's essay as reported by CNN.
The shift is agreement, not the argument
Slowing down is not a new idea. People have argued for it since 2023. What changed on 12 September 2026 is who said yes.
Amodei's essay argued the industry should deliberately slow capability gains so safety work can catch up. Then Sam Altman said OpenAI agrees and will match the first commitment. Elon Musk posted three words: "Dario is right." Demis Hassabis and Satya Nadella lined up behind the direction too.
That's the story. Not the slowdown — the alignment. The labs signaled they'll no longer be the sole arbiters of how fast AI moves.
Pacing is not pausing
Amodei was precise about what he was and wasn't proposing. Pacing, he wrote, does not mean halting model training or technical progress — it means ensuring companies take time to align and safeguard models, and let outside evaluators confirm it.
His plan has three steps, front-loaded with the one anyone can do today:
- Embed third-party evaluators inside every frontier lab, with employee-like access to pipelines — not just final models. Anthropic committed to this unilaterally.
- Coordinate across democratic nations on common safety standards.
- Coordinate globally, including with authoritarian states.
Step one is the tell. It costs nothing to agree with and needs no one's permission, which is exactly why it moved from essay to commitment in a single weekend.
The strongest objection isn't 'safety is fake'
Plenty of critics called the whole thing regulatory capture — a way to pull the ladder up on open source and freeze the current leaders in place. Former White House AI adviser David Sacks made the sharpest version, and it's worth sitting with because he agrees pacing might be right.
You're the frontier, Sacks argued. You have the duopoly. So pace it yourself — stop demanding suspended antitrust law, a hand-picked evaluator, and regulatory sign-off as the price of restraint.
You are the one setting it. The easiest way to not build super intelligence is for you to agree not to build it.
David Sacks
That's the real fault line. Not whether to slow down, but whether a coordinated, permission-seeking slowdown is safety or a cartel with better PR. The named evaluator, Meter, drew the same fire: an outside referee so entangled with lab staff and funding that critics questioned whether it counts as independent oversight at all.
The operator's angle everyone skipped
Here's the part that matters if you run a team rather than a lab.
The common read is that pacing is an economic negative — slower models, slower progress, slower returns. For enterprise adoption, the opposite may be true.
If you deal with transformation on multi-year timelines, constant model churn is a reason to not act. Why realign the whole org around a new model when it changes again in three months? Many companies quietly use that speed as an excuse for inaction — better to wait until their hands are forced.
What to copy
You don't run a frontier lab, but the move translates. When a system is changing faster than your organization can absorb, the fix isn't to match its speed — it's to set a pace your team can actually build against, and be explicit about it.
Amodei's step one is the pattern: pick the commitment that needs no one's permission, and make it unilaterally. The coordination fights come later. The credibility comes from doing the boring, immediate thing first.
The pace is finally up for negotiation — which means, for the first time, it's partly yours to set.
