Pace the Frontier Goes Cross-Partisan: Baker, Burry, Trump, and Harris React

Three days after Dario Amodei published "We Must Pace the Frontier", the debate has left the AI industry entirely. On September 14 and 15, 2026, it picked up an investor calling the whole thing hype, a cognitive scientist agreeing with him, a Palantir executive calling AI safety a political ideology, a president calling it a hoax, and a former vice president calling for Congress to legislate a slowdown. This is the story of how a lab CEO's proposal became a genuine cross-partisan, cross-industry argument — and what to actually watch for once the rhetoric settles.

A recirculating summary of X posts, titled "AI Safety Advocates Clash With Critics Over Slowdown Calls," gathered most of these reactions into one thread — it is not a primary source (it is reportedly a Grok-generated recap of the platform's own discourse), so treat it as a pointer to where to look, not as verified fact on its own. Every quote below is attributed to the individual account or statement it traces back to, not to that summary.

TL;DR

Where this picks up

explainx.ai has tracked this story since Amodei's essay landed: the original three-step plan, the first wave of industry reaction from Musk, Altman, and Hassabis, a skeptical read on whether pacing is safety or a plateau narrative, and Trump and Speaker Johnson's initial rejection of any government pause. This post does not repeat that ground — it picks up where those left off, with the specific Sept 14-15 escalation that pulled markets, researchers, and national politics fully into the argument.

By this point Sam Altman and Elon Musk had both already endorsed Amodei's proactive steps — Altman by committing OpenAI to match Anthropic's embedded-evaluator program, Musk with his one-line "Dario is right" post — a picture already detailed in explainx.ai's industry reaction roundup. What's new this round is that the argument stopped being lab-to-lab and became investor-to-investor and party-to-party.

Michael Burry and Gary Marcus: this is IPO hype

Michael Burry, the investor best known for correctly shorting the 2008 housing market, publicly dismissed the pacing wave as "self-serving hype tied to IPOs." His framing lines up almost exactly with the argument explainx.ai already laid out in Is "Pacing the Frontier" Really About Safety — Or a Plateau in Disguise?: Anthropic filed a confidential S-1 with the SEC around the same window Amodei became the industry's loudest voice on safety, and a "we could go faster but we're choosing caution" story is a considerably better pitch to public-market investors than one about data walls or expensive agentic serving costs. Burry's comment is the first time that specific critique has come from a well-known public investor rather than from tech-industry commentary.

AI researcher Gary Marcus voiced a similar skepticism — a notable pairing, since Marcus has spent years arguing current-generation LLMs are overhyped on capability, and here applies the same "don't take the industry's framing at face value" instinct to its safety claims instead. Marcus's throughline across both critiques is consistent: whatever frontier labs say publicly about their own systems — whether boasting about capability or warning about risk — deserves outside scrutiny rather than being taken as a neutral read of the technology's actual state.

Gavin Baker: pacing is a margin question, not a slowdown

Investor Gavin Baker offered the most mechanically specific read of the group. His argument: in practice, when a lab actually "paces," what changes is that it spends more compute and more calendar time on alignment work, monitoring infrastructure, and evals before shipping — not that it trains a meaningfully weaker or slower-arriving model. That's a real cost (more GPU-hours spent on non-capability work, longer time-to-release, thinner margins for whichever lab chooses to do it) but it is categorically different from the "hard technical slowdown" framing that dominates most of the public debate.

Baker's read is worth holding next to Emad Mostaque's original critique in explainx.ai's essay breakdown: Mostaque argued embedded evaluators have "minimal power" and that the only real anti-distillation lever is pausing training runs outright. Baker isn't disputing that evaluators are weak enforcement — he's making a narrower, more practical point about what the word "pacing" actually costs a lab that adopts it in good faith: money and time, not capability.

Shyam Sankar: AI safety as political ideology

Palantir CTO Shyam Sankar took the sharpest structural swing at the whole framing, arguing there is "an unseen hand pushing AI safety as a political ideology." His target isn't any specific evaluator program or piece of legislation — it's effective altruism itself, which he casts as a small group of technocrats trying to gatekeep how much technological progress everyone else is allowed to have.

This connects directly to a critique explainx.ai has already tracked from inside the industry: Chamath Palihapitiya's warning, covered in the reaction roundup, that Amodei's plan concentrates power with whichever lab writes the terms. Sankar's version widens that argument from "this specific plan favors incumbents" to "the entire intellectual movement behind AI-safety advocacy is itself an unaccountable power structure" — a materially bigger claim, and one that reframes the debate from a policy disagreement into an ideological one. Whether or not readers agree with Sankar's characterization of effective altruism, the practical effect is the same as Burry's and Marcus's contributions: another prominent, non-lab voice arguing the safety framing itself should not be taken at face value.

Trump: "I am the Hoax Buster"

President Trump extended his administration's already-stated position with considerably more force. Days earlier, Trump and Speaker Mike Johnson had rejected any government-imposed AI pause on the grounds that a domestic slowdown risks ceding ground to China. His new statement goes further than policy rejection into outright dismissal of the underlying premise:

"I am the Hoax Buster, and I'm right now breaking another Hoax — That AI is going to take over, consume, and destroy the World... This is even wilder than the RUSSIA, RUSSIA, RUSSIA HOAX, or the Global Warming Scam."

Read this as a rhetorical position, not a technical or policy analysis — the statement doesn't engage with Amodei's specific proposal (embedded evaluators, industry coordination, tiered global agreements), it rejects the entire category of concern the proposal responds to. That's a meaningfully different target than the prior Trump-Johnson statement, which at least implicitly left room for narrower oversight mechanisms like evaluator access while rejecting a pause specifically. This new framing leaves less daylight for that distinction.

Kamala Harris: the opposite pole

Former Vice President Kamala Harris posted the most detailed response of the entire round — a lengthy statement published September 15, 2026 at 2:07 AM — and it lands squarely on the opposite side of Trump's dismissal. Harris argued that frontier AI is "advancing at an alarming speed" and that the United States should "responsibly slow the pace of frontier AI development." She was explicit that this isn't a call to give up AI's benefits, citing medicine specifically as a domain where progress should continue, but argued the country needs guardrails so AI doesn't "escape human control."

Her concrete asks go further than anything either party's officials had floated in this debate so far:

• Congress passes new laws governing frontier AI development directly
• A new federal oversight and testing entity — not an existing agency repurposed, a dedicated body
• A treaty with countries including China to limit dangerous AI uses and adopt shared testing standards, while the US continues to lead in AI development

That last point is worth reading against Amodei's own framework. His essay's Step 3 — global coordination with China, up through four escalating tiers — treats pre-release testing agreements (his Level 2) as "feasible in principle" and a full pause (Level 4) as "unlikely to actually happen any time soon." Harris's proposal sits closest to that Level 2 tier: a shared testing standard, not a halt, paired with continued US leadership rather than a unilateral slowdown. It is the closest any political figure has come, on the record, to endorsing a version of Amodei's own middle tiers rather than either extreme (Trump's dismissal, or a hypothetical full pause).

One reply worth noting for its own reason: a commenter (@geek3point0) responded to Harris's statement with "This post written with AI lol. zero irony" — a piece of public skepticism about authorship and authenticity of political AI-policy statements that's worth flagging without taking a position on whether it's accurate. In an era where every major political figure has AI-drafting tools available and every long-form policy statement invites that exact question, it's a small but telling data point about how political AI messaging is now read by default with suspicion, regardless of the statement's content.

Reading all five reactions together

Line these up and a pattern emerges that's more informative than any single reaction: everyone is looking at the same proposal through their own professional lens.

• Investors (Burry, Baker) read it as a market and cost story — either as investor-relations theater (Burry) or as a real but modest margin hit (Baker). Neither treats it primarily as a safety question.
• A researcher (Marcus) applies his usual "don't trust the industry's self-description" skepticism, consistently, whether the claim is about capability or about caution.
• An industry executive (Sankar) reframes it as an ideological power struggle, not a technical or policy dispute at all.
• Politicians (Trump, Harris) split along the oldest fault line in this entire debate — whether the underlying risk is real enough to legislate — but for the first time, both sides of that split are now on the record with a fully developed public position, not just one side rejecting a hypothetical.

None of these readings are mutually exclusive, and none of them individually settle whether Amodei's original proposal is sincere safety work, market positioning, or something in between — a question explainx.ai has already examined directly in Is "Pacing the Frontier" Really About Safety — Or a Plateau in Disguise?. What's changed is the size of the audience arguing about it.

What to actually watch next

For builders and researchers who care about outcomes rather than rhetoric, three concrete things determine where this goes from here, not any individual tweet or statement:

1. The September 24, 2026 Trump-Xi AI governance meeting. This is the first real test of Amodei's hardest step — global coordination including China. Whatever comes out of it (or doesn't) will do more to settle the pacing debate than any domestic political statement, since it's the one venue where an actual US-China agreement, even a narrow one, could be negotiated.
2. Whether "pacing" ends up meaning compute/eval spending (Baker's read) or legislation (Harris's read) or nothing at all (Trump's read). These are three genuinely different outcomes with different practical effects on release timelines, compliance costs, and what a builder can expect from frontier labs' API behavior over the next year.
3. Whether OpenAI's, DeepMind's, or xAI's evaluator programs actually launch with real operational terms, the same signal explainx.ai flagged in the prior reaction roundup — since that remains the most concrete, checkable evidence of whether any of this translates into an actual practice rather than a talking point.

Related reading

• Update — September 15, 2026: The same day, a separate viral thread from Kevin Bass argued METR and the AI-safety funding ecosystem can't independently evaluate Anthropic because of shared donor money — a distinct but related independence question. The "Anthropic Network" claim, fact-checked →
• Dario Amodei Wants to "Pace the Frontier" — Here's the Actual Plan
• Musk Backs Amodei, Altman Commits OpenAI to Match: The Pace the Frontier Reaction
• Is "Pacing the Frontier" Really About Safety — Or a Plateau in Disguise?
• Trump and Speaker Johnson Reject an AI Pause — "Keep the China Lead" Wins
• Jack Dorsey's "Open the Frontier" Answers Amodei — With Open Weights
• Anthropic Files Confidential S-1 With the SEC: AI Safety Leader Eyes IPO
• What Is an Embedded Evaluator in AI Safety?
• G20 Backs the "Carolina Principles": Light-Touch AI Rules, China Included

This post reflects public statements circulating on X and covered in secondary summaries as of September 15, 2026, including a widely circulated Grok-generated recap titled "AI Safety Advocates Clash With Critics Over Slowdown Calls." Quotes attributed to individuals are drawn from their own posts or statements; the recap itself is not treated as a primary source. Check each figure's own account or an official statement before treating any single line here as a verbatim, final record — this is a fast-moving, multi-platform news cycle and phrasing may be clarified or disputed after publication.