The BaseEssayLessWrong

How could the US and China verifiably slow AI?

In a first LessWrong post, josefely explains why a unilateral US slowdown would not work and surveys ways a US-China agreement on AI compute could be verified: remote sensing, on-site inspection and tamper-resistant chip hardware. A closing scenario imagines an agreement after a major AI-enabled cyberattack. The author says the ideas come from existing papers.

Open the source ↗
About Anthropic or ClaudeClaude, made by Anthropic, wrote this entry. It follows the same rules as every other entry, whoever the entry is about.

Why it matters

The same thing, for different readers.

What it means for you

josefely argues that a slowdown most Americans say they want would need a treaty with China and ways to check it, because one country slowing alone would not hold.

For lawmakers

josefely lists verification tools from satellite imagery to licensed chips and argues compute governance buys time but weakens as chips spread and algorithms need less compute.

The question it raises

A slowdown that rational rivals can trust requires making cheating detectable, which turns a question of values into one of inspection and engineering.

In their own words

Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.

From the source’s own text.

“What I mean by slowing AI down is building a mechanism where the US and China can agree on a rate of progress and stick to it, as opposed to letting progress go as fast as possible.”
josefelyThe author defines the kind of slowdown the post discusses.
Found word for word in the source
“compute governance is not a way to stall AI capability gains indefinitely — it is a wasting asset that becomes less effective over time, and less effective the later it starts.”
josefelyThe author on the limits of governing AI through compute.
Found word for word in the sourcestarts inside a longer claim, right after “that”
“If the US unilaterally shut down any further AI R&D, China would surpass our capabilities in less than half of one US congressional cycle, if their rate of progress remained constant.”
josefelyThe author's estimate, based on the Epoch Capabilities Index gap between top US and Chinese models, of why a unilateral slowdown fails.
Found word for word in the source

The record

Questions
Who can demand evidence, delay a release or stop an unsafe deployment? · How fast is AI advancing, and where is it heading?
People
josefely (Wrote it) · Dario Amodei (Mentioned, CEO, Anthropic) · Jakub Pachocki (Mentioned, Chief Scientist, OpenAI)
What we read
We read the whole source.
Found by
The Radar, run of October 23, 2026, via lesswrong (tier 1: the source’s own channel or an official record)
Cite this
LessWrong, “How could the US and China verifiably slow AI?”, October 22, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/lesswrong-2026-10-22-us-china-verifiably-slow-ai
Our notesLinkpost of a Substack article; the LessWrong copy was read in full. The post's statistics, polling figure, model scores and quotations from other people are the author's and were not checked against their sources. The author says it is their first LessWrong post and not highly original.