'Stop Pretending This Is Altruism': Sacks Calls Amodei's Pacing Plan 'Regulatory Capture' as the Backlash Arrives

Within hours of Dario Amodei’s “We Must Pace the Frontier” essay going live on September 12, the AI industry’s most powerful voices lined up behind it. Elon Musk posted “Dario is right.” Sam Altman agreed OpenAI would adopt the same evaluator-access model. Hugging Face announced an Open Alignment Initiative to get a seat inside the embedded-evaluator program Amodei proposed. For a moment, it looked like the first genuine cross-lab consensus on slowing down.

Then the backlash arrived — and it came from the three directions that matter most: the White House, Meta, and the open-source movement Musk himself claims to champion.

Sacks: ‘Regulatory capture’ by incumbents

White House PCAST chair David Sacks used an X post on Saturday, September 12 to reject Amodei’s framework outright, and his choice of words was pointed: “regulatory capture.”

Sacks told Anthropic and OpenAI to “go ahead” and pace themselves without waiting for Washington to sanction it. His argument, reconstructed across multiple reports of the post, rests on three pillars:

1. The cartel question. Amodei’s plan asks democracies to coordinate frontier-AI slowdowns — which, Sacks noted, would seemingly require antitrust relief “to form a cartel.” Coordinated capability throttling among the labs already in the lead looks less like independent safety caution and more like collusion that locks in whoever is ahead.

2. The motivation question. Sacks pushed the labs to “stop pretending the motivation to slow down is purely altruistic,” pointing to their product-liability exposure. If pacing the frontier also happens to reduce legal risk for the companies proposing it, the safety framing deserves scrutiny.

3. The unilateral option. Nothing stops Anthropic and OpenAI from slowing down on their own, today, without a government-blessed coordination framework. If the argument is purely about safety, why does the plan need Washington’s antitrust waiver at all?

This is not a new position for Sacks — he accused Anthropic of running “a sophisticated regulatory capture strategy based on fear-mongering” as far back as October 2025, and renewed the charge in an August 2026 spat over Amodei’s “DMV for AI” framing. But Saturday’s post is the first time the White House has directly answered the pacing proposal itself, and the first time the disagreement has been this explicit: the administration’s position is that the labs do not need — and should not get — a coordination mechanism.

Wang: alignment as ‘the gating factor for scaling’

The same day, Meta Chief AI Officer Alexandr Wang posted a statement that reads as adjacent to Amodei’s proposal but pointedly does not join it:

“Alignment is fundamental to delivering personal superintelligence for everyone. People need agents they can trust to reliably do what they ask. MSL is rapidly scaling up the share of our efforts that goes into alignment as our models become more powerful. We do believe alignment can be the gating factor for scaling as we get closer to the frontier.”

The substantive claim is the phrase “gating factor.” A gating factor is whatever constraint actually determines your ceiling, regardless of what else you have. Wang is saying that as Meta Superintelligence Labs (MSL) approaches frontier capability, alignment confidence — not GPU supply, not researcher headcount, not data — could become the bottleneck that caps output.

That is a stronger and more falsifiable claim than a generic “we care about safety” statement, because it implies a testable prediction: at some point, MSL should visibly slow or delay a release, citing alignment concerns specifically. That is the concrete thing to watch for.

The statement is also deliberately uncoordinated. As one reply noted, “I don’t see the word pacing anywhere.” Wang’s post is about MSL’s own internal gating decision — not a call for industry-wide standards the way Amodei’s essay explicitly is. Meta is endorsing the premise (alignment can slow scaling) while declining the mechanism (coordinated pacing with rivals).

The post did not go unanswered. A reply quoting Meta whistleblower Jason Sattizahn’s September 2025 allegations — “We researchers were directed to write reports to limit risk to Meta” — put the credibility question bluntly: Meta’s track record of internal reviews does not obviously support the claim that alignment work will be allowed to gate a release when it is inconvenient. A company can hold a correct alignment philosophy on paper while internal incentives — pressure to ship, pressure to limit reputational risk — override it in practice. Nothing in Wang’s statement resolves that prior allegation.

Musk: ‘Nothing can shut down open source’

The strangest turn came from Musk himself. In a separate September 13 thread, Kevin Bass noted that Musk’s endorsement of Amodei’s pacing advocacy was “concerning” given that Musk is “overwhelmingly the largest tech donor to the House GOP Super PAC” — and given what Congress is about to do. Musk’s reply was four words:

“Nothing can shut down open source.”

As a narrow technical claim, this is true: once model weights are publicly released and mirrored across enough servers and torrents, no single government action can retroactively delete every copy. It is a structural fact about file distribution.

But it does not address what pending legislation would actually regulate — which is less about deleting existing weights and more about restricting who can legally use, fine-tune, host, or commercially deploy them going forward, and what compliance burden falls on new open-weight releases. As one reply put it: “Political oversight and regulations can absolutely shut it down. Leaving the technology only for frontier labs and criminals.”

That asymmetry is the actual policy fight. Compliance costs — mandatory safety testing, licensing for training runs above compute thresholds, liability for downstream misuse — fall disproportionately on developers who follow the law. A well-funded frontier lab can absorb the overhead; a solo developer or academic fine-tuning an open-weight model may not. An actor with no intention of complying with US law is unaffected either way. The practical effect critics worry about is a regulatory landscape that filters out the middle: capability concentrated among large compliant labs and unregulated bad actors, with fewer legitimate independent builders able to operate in between.

The e/acc flank: ‘Don’t join the pacing cartel’

The effective accelerationist movement supplied its own vocabulary for the fight. Beff Jezos replied to Wang: “Don’t join the pacing cartel please.” “Pacing cartel” is the e/acc term for what happens when frontier labs coordinate on safety-driven slowdowns — the argument being that once multiple labs agree to throttle capability growth together, the effect looks like collusion dressed in safety language, locking in whoever is already ahead.

The same suspicion runs through Emad Mostaque and Chamath Palihapitiya’s earlier critiques of Amodei’s plan: a pacing framework concentrates power with whoever writes its terms. Tech writer Xe Iaso’s September 12 satire — “Everyone should slow down AI development except for me,” imagining a fictional lab CEO calling for rivals to pause so his own lab can catch up — hit 318 comments on Hacker News within hours, crystallizing the same doubt in joke form.

Congress moves — under pressure

The backdrop to all of this is legislative. Axios reported on September 11 that House Democrats are urging Speaker Mike Johnson to cancel the House’s scheduled recess to pass AI safeguards, with lawmakers warning that Congress may be leaving dangerous gaps unaddressed. The Independent reported on September 3 that Johnson had already cancelled final weeks of House votes before the midterms — meaning the legislative calendar is extraordinarily compressed exactly when the pacing debate has reached its peak.

The result is a debate that usually lives in essays and reply threads suddenly carrying actual legislative stakes. If Congress writes open-weight rules in haste, the open-source counter-argument — that regulation filters out the compliant middle — stops being theoretical.

Where this leaves the pacing consensus

The 24 hours after Amodei’s essay produced something genuinely unusual: the CEOs of Anthropic, OpenAI, xAI, and the largest open AI platform endorsing the same slowdown framework. The 24 hours after that produced its mirror image: the White House’s AI czar calling it a cartel strategy, Meta endorsing the premise while rejecting the mechanism, Musk pivoting to an open-source absolutism that contradicts the coordination his endorsement implied, and Congress under pressure to legislate before the calendar runs out.

The pacing consensus, it turns out, was the easy part. The fight over who gets to set the terms — and what happens to open weights when governments start writing them — is just beginning.