Every internet argument ends in an alignment chart eventually, and the pacing debate has earned one. The classic grid has two axes, Lawful to Chaotic and Good to Evil. We kept the first and swapped the second, because nobody in this fight wants the bad outcome. They disagree about two things: who should referee the pace of frontier AI, and whether to slow it down at all.
The Frontier Alignment Chart
The leading voices from Pace the Frontier, placed by who should referee the pace and where they stand on slowing the frontier. Tap a face for the reasoning.
Lawful
The law referees
Neutral
The labs referee
Chaotic
Nobody referees
Pace
For pacing or slowing the frontier, up to a full stop.
Neutral
Ambivalent, conditional without a clear side, or no public position yet.
Dario AmodeiWrote the essay: slower capability growth, embedded evaluators and coordinated standards.
Demis HassabisAgrees with the direction and points to his proposal for a frontier standards body.
Ilya SutskeverSigned the July letter, on condition that it is international and well implemented.
Gillian HadfieldBuild the option to pace, with publicly overseen verifiers, and keep models from building models.
Max TegmarkA prohibition until there is safety consensus and public buy-in, with safety evidence before deployment.
Eliezer YudkowskyWants large training runs halted indefinitely and worldwide; six months was never enough.
Connor LeahyProposed a binding international moratorium on frontier runs, enforced by a compute cap.
Barack ObamaCalls the labs slowing down a necessary first step, with the public involved in deciding.
Bernie SandersAnnounced legislation for an immediate pause and a permanent ban on superintelligence.
Pope Leo XIVAsks for regulation, human oversight and, at times, a slower pace of adoption, in an encyclical.
Sam AltmanBacks pacing and commits OpenAI to independent evaluators with employee-level access.
Andrej KarpathyWants the industry to come together and implement the proposal.
Neel NandaWelcomes verification, but wants a specific agreement and real action.
Josh EngelsPacing has to buy alignment time to catch up.
Sholto DouglasArgues pacing costs his own lab and gives challengers time to catch up.
John SchulmanWants labs to design voluntary coordination before governments step in.
Drake ThomasBacks the letter and asks for interpretability, audits and biosecurity work.
Geoffrey HintonSigned the statement asking for scientific consensus and public buy-in before superintelligence.
Yoshua BengioSigned the same statement, and backs the July pacing letter as well.
David KruegerA signatory of the prohibition-until-safe statement.
Daniel KokotajloReal pacing should slow the leaders more than the challengers.
Elon Musk“Dario is right.” Three words of support, with no commitment attached.
Gary MarcusNo broad slowdown yet. Instead, an enforceable recall of unreliable internet-connected agents.
Satya NadellaEndorses Hassabis’s standards body; no verified reaction to the pacing essay itself.
Alexandr WangSays alignment could become the gate on scaling, without endorsing coordinated pacing.
Ethan MollickWatching METR become a de facto standard setter, without taking a side.
Sundar PichaiNo public position on the pacing proposal yet. True neutral by default.
Sergey BrinNo public position on the pacing proposal yet either.
Mustafa SuleymanAnswered with a code of conduct (no race to a runaway superintelligence), not a view on pacing.
Clément DelangueLaunched an open alignment effort: safety through transparency, not a shared slowdown.
roon (@tszzl)There are many options between open weights and no sharing at all. Commits to none of them.
Donald TrumpThe only guardrail AI needs is a strong president. Whoever wins AI wins.
Xi JinpingBeijing’s reply rejects curbs on China’s AI and calls on all parties to work on global governance.
Alex KarpWould favor a pause if there were no adversaries. There are, so no pause.
David SacksLabs are free to slow down on their own. No antitrust carve-out, no new framework.
Mark ZuckerbergEvery lab has the responsibility and the incentive to pace itself. No coordination needed.
Jensen HuangPro-development: reportedly called AI safe and good for jobs, growth and national security.
Chamath PalihapitiyaArgues the proposal would suppress open source and concentrate power.
Jason CalacanisIf you want safety, you want disclosure, and open source is the strongest kind.
Yann LeCunOpposed the six-month moratorium in 2023.
Andrew NgOpposed the six-month moratorium in 2023, alongside LeCun.
Jack DorseySupports open releases and scrutiny, not limits negotiated by incumbent labs.
Jürgen SchmidhuberCalls recursive self-improvement ML’s most rewarding goal and welcomes the startups chasing it.
Tap a face to see why it sits where it does. Near a line means a borderline call.
Lawful:
Governments, treaties, regulators and public standards bodies set the pace.
Neutral:
Labs coordinate among themselves: voluntary deals, evaluators, codes of conduct.
Chaotic:
No central referee: open weights, markets and individual calls decide.
Pace:
For pacing or slowing the frontier, up to a full stop.
Neutral:
Ambivalent, conditional without a clear side, or no public position yet.
Accelerate:
Against a coordinated slowdown.
Good and Evil are retired here: the vertical axis is each person’s stance on pacing, and nobody on this chart is a villain for wanting a different speed. Placements are our editorial reading of the statements recorded in the report, not labels anyone chose for themselves.
Every placement, with the reasoning
Lawful Pace
Dario Amodei · Wrote the essay: slower capability growth, embedded evaluators and coordinated standards.
Demis Hassabis · Agrees with the direction and points to his proposal for a frontier standards body.
Ilya Sutskever · Signed the July letter, on condition that it is international and well implemented.
Gillian Hadfield · Build the option to pace, with publicly overseen verifiers, and keep models from building models.
Max Tegmark · A prohibition until there is safety consensus and public buy-in, with safety evidence before deployment.
Eliezer Yudkowsky · Wants large training runs halted indefinitely and worldwide; six months was never enough.
Connor Leahy · Proposed a binding international moratorium on frontier runs, enforced by a compute cap.
Barack Obama · Calls the labs slowing down a necessary first step, with the public involved in deciding.
Bernie Sanders · Announced legislation for an immediate pause and a permanent ban on superintelligence.
Pope Leo XIV · Asks for regulation, human oversight and, at times, a slower pace of adoption, in an encyclical.
Neutral Pace
Sam Altman · Backs pacing and commits OpenAI to independent evaluators with employee-level access.
Andrej Karpathy · Wants the industry to come together and implement the proposal.
Neel Nanda · Welcomes verification, but wants a specific agreement and real action.
Josh Engels · Pacing has to buy alignment time to catch up.
Sholto Douglas · Argues pacing costs his own lab and gives challengers time to catch up.
John Schulman · Wants labs to design voluntary coordination before governments step in.
Drake Thomas · Backs the letter and asks for interpretability, audits and biosecurity work.
Geoffrey Hinton · Signed the statement asking for scientific consensus and public buy-in before superintelligence.
Yoshua Bengio · Signed the same statement, and backs the July pacing letter as well.
David Krueger · A signatory of the prohibition-until-safe statement.
Daniel Kokotajlo · Real pacing should slow the leaders more than the challengers.
Chaotic Pace
Elon Musk · “Dario is right.” Three words of support, with no commitment attached.
Lawful Neutral
Gary Marcus · No broad slowdown yet. Instead, an enforceable recall of unreliable internet-connected agents.
Satya Nadella · Endorses Hassabis’s standards body; no verified reaction to the pacing essay itself.
True Neutral
Alexandr Wang · Says alignment could become the gate on scaling, without endorsing coordinated pacing.
Ethan Mollick · Watching METR become a de facto standard setter, without taking a side.
Sundar Pichai · No public position on the pacing proposal yet. True neutral by default.
Sergey Brin · No public position on the pacing proposal yet either.
Mustafa Suleyman · Answered with a code of conduct (no race to a runaway superintelligence), not a view on pacing.
Chaotic Neutral
Clément Delangue · Launched an open alignment effort: safety through transparency, not a shared slowdown.
roon (@tszzl) · There are many options between open weights and no sharing at all. Commits to none of them.
Lawful Accelerate
Donald Trump · The only guardrail AI needs is a strong president. Whoever wins AI wins.
Xi Jinping · Beijing’s reply rejects curbs on China’s AI and calls on all parties to work on global governance.
Alex Karp · Would favor a pause if there were no adversaries. There are, so no pause.
Neutral Accelerate
David Sacks · Labs are free to slow down on their own. No antitrust carve-out, no new framework.
Mark Zuckerberg · Every lab has the responsibility and the incentive to pace itself. No coordination needed.
Jensen Huang · Pro-development: reportedly called AI safe and good for jobs, growth and national security.
Chaotic Accelerate
Chamath Palihapitiya · Argues the proposal would suppress open source and concentrate power.
Jason Calacanis · If you want safety, you want disclosure, and open source is the strongest kind.
Yann LeCun · Opposed the six-month moratorium in 2023.
Andrew Ng · Opposed the six-month moratorium in 2023, alongside LeCun.
Jack Dorsey · Supports open releases and scrutiny, not limits negotiated by incumbent labs.
Jürgen Schmidhuber · Calls recursive self-improvement ML’s most rewarding goal and welcomes the startups chasing it.
Across: who referees.Lawful wants the pace set by governments, treaties, regulators or a public standards body. Neutral wants the labs to sort it out among themselves, with voluntary agreements, embedded evaluators and codes of conduct. Chaotic wants no central referee at all: open weights, markets and individual calls.
Down: where they stand on pacing.Pace is for slowing the frontier, from embedded evaluators all the way to a full stop. Neutral is ambivalent, conditional without picking a side, or hasn't said anything yet. Accelerate is against a coordinated slowdown.
Every placement is our reading of the statement the Pace the Frontier report recorded for that person, with their source one tap away. Nobody picked their own square, and several people would argue for the one next door. That's half the fun of an alignment chart.
Pace is the biggest row (27 people), and Neutral Pace is the crowded square (15): lab leaders and researchers who want the labs to slow down together, plus the scientists who signed for a stop until there is consensus.
Lawful Pace runs from the essay to the encyclical (10): Dario Amodei and Demis Hassabis share it with Bernie Sanders, Eliezer Yudkowsky and Pope Leo XIV.
Chaotic Accelerate comes next (11): the open-source camp and the investors who read the proposal as regulatory capture.
The Neutral row is where the open-source pragmatists live (17): Hugging Face wants open verification rather than a shared speed limit, and Sundar Pichai and Sergey Brin haven't said anything yet.
Lawful Accelerate is the heads-of-state corner. Donald Trump and Beijing both reject a slowdown and both see the state as the guardrail, even if they agree on little else.
