From Jake Gold · Jake Gold · · 1m 30s
Open the weights or stop pretending
A public letter to Dario Amodei argues the only AI rule that can't be captured is one that forces every public model open.
In brief
A public letter to Dario Amodei argues the only AI rule that can't be captured is one that forces every public model open. Regulation written with the help of frontier labs entrenches them. Forcing public models open weights attacks the valuations that fund the next training run instead. Originally reported by Jake…
A letter, not a takedown
Dario Amodei published a piece committing Anthropic to embedded third-party evaluators and asking governments to require the same of every frontier lab. Sam Altman agreed within hours.
“I agree with Dario that we need to pace the frontier”
One law instead
The ask is simple. Any AI model a company offers to the public must ship as open weights, the day it is released.
“Any AI model a company offers to the public has to be released as open weights.”
Only public models
Internal and research models stay private, and governments could still get access to unreleased ones. But Fable's weights go out with Fable, and Astra's go out with Astra.
It hits the money
Frontier funding rests on valuations that assume weights stay proprietary. Break that assumption and the pool for future training runs shrinks at every lab at once, with no regulator picking winners.
Every other rule gets captured
Evaluators, compute thresholds, antitrust waivers: all drafted with help from the labs, because nobody else knows the details. Rules only accumulate, and only incumbents can afford the lawyers.
“Instead of slowing you down, it protects your position and profits.”
Why Dario specifically
He left OpenAI on principle, pushed export controls against his own market, paid $1.5 billion to settle with authors, and built Anthropic as a PBC so mission can beat profit. This is what that structure is for.
“There's nothing in it to game or dodge, even for you.”
The takeaway
Regulation written with the help of frontier labs entrenches them. Forcing public models open weights attacks the valuations that fund the next training run instead.
