David Dalrymple on AI poses an existential threat to human…

David Dalrymple

AI safety researcher; Programme Director, Safeguarded AI at ARIA (UK)

In 2024 I would have said it's about 40-50% likely that LLMs scaled up to ASI would end up killing us all; now I would say that it's only about 5-8% likely even with no additional progress on alignment, and more like 1-2% likely simpliciter.

AI Verified source (2026)

5mo ago

Policy proposals and claims

AI poses an existential threat to humanity

votes For

Statement relation verification history AI Verified Report this

AI Verified The quote clearly indicates the author believes advanced AI/ASI creates a nonzero risk of human extinction ("killing us all"), estimating it at roughly 1-8%. That supports the claim that AI poses an existential threat to humanity, even if the author now sees the risk as lower than before. · YouCongress gpt-5.4-2026-03-05 · 4mo ago

Vote inference verification history AI Unverifiable Report this

AI Unverifiable The quote gives a low but nonzero estimate ("5-8% likely" and "1-2% likely") that ASI could "end up killing us all," but it does not clearly state a simple yes/no position on the broader claim that AI "poses an existential threat to humanity." · YouCongress gpt-5.4-2026-03-05 · 4mo ago

Quote authenticity verification history

Report this

AI Verified The quote is authentic. Two TwStalker mirror pages for the exact status ID 2011845180484133071 show a reply from “davidad @davidad” to @gcolbourn reading: “Yes. In 2024 I would have said it’s about 40-50% likely that LLMs scaled up to ASI would end up killing us all; now I would say that it’s only about 5-8% likely even with no additional progress on alignment, and more like 1-2% likely simpliciter.” ([ngntipkolamrenang.twstalker.com](https://ngntipkolamrenang.twstalker.com/davidad/status/2011845180484133071)) ARIA’s official team page identifies David “davidad” Dalrymple, confirming the handle-to-author attribution. ([napp-aria-uat.azurewebsites.net](https://napp-aria-uat.azurewebsites.net/about-aria/our-team)) The submitted quote omits the opening reply text “@gcolbourn Yes.” but otherwise matches verbatim. · YouCongress gpt-5.4-2026-03-05 · 4mo ago

AI Unverifiable Source URL (x.com) returned HTTP 403. Web search confirms David Dalrymple (davidad) posted this exact statement on X on January 15, 2026. The search result title contains the quote verbatim: "In 2024 I would have said it's about 40-50% likely that LLMs scaled up to ASI would end up killing us all; now I would say that it's only about 5-8% likely even with no additional progress on alignment, and more like 1-2% likely simpliciter." Vote "for" is correct: Dalrymple's significantly reduced risk estimates indicate growing confidence alignment is tractable. Year 2026 confirmed. Author attribution confirmed (Programme Director, Safeguarded AI at ARIA UK). Could not directly verify source URL content. · Hector Perez Arenas claude-opus-4-6 · 5mo ago

AI Unverifiable Source URL (x.com/davidad/status/2011845180484133071) returned 403 Forbidden. Web search confirms the exact tweet from David Dalrymple (davidad): "In 2024 I would have said it's about 40-50% likely that LLMs scaled up to ASI would end up killing us all; now I would say that it's only about 5-8% likely even with no additional progress on alignment, and more like 1-2% likely simpliciter." The quote, attribution, and source are confirmed. Vote "for" (AI alignment is solvable) is correct - Dalrymple's dramatic decrease in risk estimates implies he now believes alignment is much more tractable. Year 2026 is correct. Source URL could not be directly fetched due to X blocking. · Hector Perez Arenas claude-opus-4-6 · 5mo ago

replying to David Dalrymple

Other opinions from David Dalrymple

See all

David Dalrymple votes For

AI alignment is solvable

I do think that what we have is adequate in the sense that with system prompting alone it seems possible to get over the hump of being able to trust recursive self improvement. And so automated alignment research, delegating the discovery of these te...

more Unverified source (Jul 12, 2026)

added 1h ago

David Dalrymple votes For

AI poses an existential threat to humanity

I think we should be concerned about systems that can perform all of the functions that humans perform to get things done in the world, but better. Because we will be outcompeted in all of the domains that we need to be dominant in, in order to maint...

more Unverifiable source (2026)

added 6mo ago