The BaseEssayAI Alignment Forum

Where is all the AI safety research uplift?

Buck, writing on the AI Alignment Forum, takes METR's estimate of a 1.7x speedup at Anthropic to imply far larger per-researcher uplift, then asks why outside safety researchers see only about 2-3x. Buck lists possible explanations and says the answer should show soon.

Open the source ↗3 quotations checked, word for word ↓
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.

For researchers

Buck reports that Redwood's empirical safety work is only about 2-3x faster with current AI, well below what METR's Anthropic figure would suggest, and lists reasons it might differ.

The question it raises

If AI speeds capabilities work more than safety work, the hope of using AI to make AI safe is undermined; Buck says the coming months will test it.

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.

“If this is the case, then AIs speed up AI safety research as much as AI capabilities research, we might expect Anthropic safety researchers to be working 10x faster, and external safety researchers to be working 5x faster now and 10x faster soon.”
BuckBuck's inference from METR's report and Anthropic staff estimates of uplift.
Found word for word in the source
“I don't think our empirical research has been going 5x faster since we got access to Fable/Opus 5.5; I would have guessed more like 2-3x speedup. Why is this?”
BuckBuck on the Redwood team's experience, which motivates the post.
Found word for word in the source
“At the very least, if we don’t see huge speedups on the kind of AI safety work that is most amenable to huge speedups, I’m going to think the researchers doing that work are doing something wrong.”
BuckClosing challenge to external safety researchers.
Found word for word in the source

The record

Questions
What happens when AI helps build more capable AI? · Are AI companies putting stronger safeguards into practice?
People
Buck (Wrote it) · leogao (Speaks, commenter)
What we read
We read the whole source.
Found by
The Radar, run of November 10, 2026, via ai-alignment-forum (tier 1: the source’s own channel or an official record)
Cite this
AI Alignment Forum, “Where is all the AI safety research uplift?”, November 9, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/alignmentforum-2026-11-09-where-is-safety-research-uplift
Our notesAuthor shown as 'Buck'; surname id assumed from the Redwood site footer of a related post, not from this page. Verify. Estimates are the author's informal numbers.