DeepSeek Engineer: Anthropic Getting AGI Is 'Hitler With the A-Bomb'

The Post That Broke the Silence

On Monday, DeepSeek engineer Liu Shengyu did something no one in the Chinese AI industry had done quite so bluntly: he likened Anthropic getting to AGI first to "Hitler with the A-bomb."

The post, made on WeChat and covered September 15 by the South China Morning Post, pulls no punches. Liu said he did not trust Anthropic or OpenAI to make cutting-edge AI "open and affordable," and contrasted their closed-source model with DeepSeek's own open-weight approach. His core fear: if a single closed lab controls the most advanced AI, "social mobility would become increasingly difficult."

Then came the historical analogy that made the story go wide: Liu compared the prospect of Anthropic mastering AGI to Adolf Hitler and Nazi Germany acquiring nuclear weapons before the Allies.

Three days later, the essay has become the sharpest Chinese-industry response yet to Anthropic CEO Dario Amodei's Saturday essay calling on the US to "pace" frontier AI development — and it landed squarely in the middle of a week where AI safety is set to land on the agenda of a Xi-Trump meeting expected on September 24.

What Liu Actually Said

Liu Shengyu is not a random commenter. He is the engineer behind DeepSeek's V4.1 models — the same V4 series that, according to SemiAnalysis, just set a new token-throughput record on Nvidia's Vera Rubin hardware at 7x Blackwell's tokens per megawatt. That's the technical credibility behind the political statement.

Liu's argument has three layers:

1. Concentration risk. A single closed-source company controlling AGI means the benefits — and the steering — are centralized. "Social mobility would become increasingly difficult," he wrote. That is theconcern OpenAI and Anthropic say they share, but Liu sees their closed-weight approach as a contradiction.

2. Open source as counterbalance. Liu framed staying at DeepSeek as a bet on open source as the thing that keeps frontier AI from becoming a monopoly. "I especially don't want Anthropic to master the most advanced artificial intelligence or AGI," he wrote.

3. The Hitler analogy. The nuclear-weapons comparison is the part that got the coverage, but it is the most freighted. It casts Anthropic's pacing call not as safety-minded restraint but as a bid to lock in a position — the same way a nuclear monopoly would.

The Context: Amodei's Essay and the Week It Landed In

Anthropic CEO Dario Amodei published his essay on Saturday, September 13, calling on frontier AI companies to "pace" development because capabilities were advancing faster than safeguards. Sam Altman and Elon Musk both backed the call to slow the race.

Amodei also singled out China, arguing a Chinese lead in AI would pose a "grave danger" to the US and the world, and called on Washington to maintain export controls on advanced chips and chipmaking equipment.

Liu's essay is a direct rebuttal to that framing — published the same day China's Foreign Ministry pushed back. Spokesperson Guo Jiakun called Amodei's remarks "fearmongering, confrontation and vicious competition" that would "disrupt the process of global AI governance." The state-run Global Times separately accused Anthropic of engineering a "silent AI Cold War."

By Monday, Trump was on Truth Social calling Amodei a "perfect little angel" and declaring the only AI guardrail the country needs is "a STRONG AND SMART (High IQ!) PRESIDENT." His administration, he said, has "tremendous CRIMINAL and REGULATORY power" over AI firms.

The Market Reaction Was Real

This is not just rhetorical. The debate moved markets:

• Nvidia fell 3.4% on Monday.
• Micron lost more than 5%.
• Broadcom and AMD each dropped more than 4%.
• The Nasdaq Composite closed 0.56% lower and the S&P 500 fell 0.48%.
• Hong Kong's Hang Seng AI Index slipped 1.65% on Tuesday, extending Monday's 3.3% drop.
• Z.ai (Zhipu AI) closed down 5.69% and MiniMax fell 7.05% on Tuesday.

The PyTorch of it: a single engineer's social media post, combined with an essay from an American CEO and a presidential response, moved billions in market capitalization across two continents in 48 hours.

The Xi-Trump Meeting Changes the Stakes

The expected September 24 meeting between President Xi and President Trump is being positioned as potentially the most consequential AI-policy dialogue since the chip export controls began. AI safety is set to surface as a key point of discussion.

Liu's essay landed in the exact window where both sides are calibrating their public positions ahead of that meeting. The timing is not accidental — it is a signal.

The Chinese Industry Is Not unified

Liu's post is the loudest voice, but it is not the only one. The Dongbuyaqiao Research Institute, an independent Chinese research platform, argued Amodei's pacing call could be "an attempt to gain greater control over industry rules before they are fully established." Du Hengfeng, a commentator for National Business Daily, wrote that "direct commercial returns are the deeper motivation" behind the US tech giants' safety calls.

That is a coordinated-anxiety reading: the Chinese industry sees the pacing argument as a strategic move dressed up as safety, not as genuine risk management.

What Anthropic and DeepSeek Had to Say

Neither Anthropic nor DeepSeek immediately replied to emailed requests for comment on Tuesday, according to SCMP. That silence is itself informative — Anthropic is days away from what Axios reports could be a US$2 trillion IPO targeting valuation, and a public fight with a Chinese engineer does not help the narrative.

DeepSeek, meanwhile, has every incentive to let Liu's post stand as a proxy argument for its open-weight philosophy.

The Open-Source vs Closed-Source Fault Line

This is the real story underneath the Hitler headline. The US-China AI debate has been cast as a national-security competition — chips, compute, export controls. Liu's essay reframes it as something more fundamental: who gets to build on top of the most powerful models, and who does not.

DeepSeek's V4.1 models are open-weight. Anthropic's Claude and OpenAI's GPT series are closed. That is not a trivial difference — it determines whether a startup in Nairobi, a researcher in Buenos Aires, or a developer in Shenzhen can run the frontier model locally, fine-tune it, and build on it without permission.

Liu's argument is that a closed-source AGI monopoly is more dangerous than a fast-moving open one. That is a legitimate position that deserves to be debated on its own terms, not dismissed because of the historical analogy.

Honest Caveat

Liu's post is a social media essay, not a peer-reviewed safety analysis. The Hitler analogy is historically loaded and will overshadow the substantive argument for many readers. And the market reaction — while real — cannot be cleanly attributed to one essay; it was a confluence of Amodei's Saturday piece, Trump's response, and broader AI-pacing anxiety.

But the fact that the essay landed with this much force — and that SCMP, a mainstream outlet, covered it as a lead tech story — tells you something: the open-source counterargument to the pacing movement has found its most vivid voice yet.

The Numbers Behind the Story

The Takeaway

Liu Shengyu's claim that Anthropic getting AGI is "like Hitler with the A-bomb" is the most provocative framing of the open-source vs closed-source AI debate yet from someone inside the Chinese frontier industry. Whether you find the analogy illuminating or inflammatory, it has moved markets, shaped coverage, and landed in the middle of the most consequential US-China AI policy week of the year.

The more durable argument underneath the headline — that a single closed lab controlling AGI is a concentration risk, and that open weights are a counterbalance — is one the pacing movement has not yet answered directly. This week's market reaction suggests it is an argument people are ready to hear.

Sources: South China Morning Post, September 15, 2026; Anthropic CEO essay, September 13, 2026; Trump Truth Social posts, September 14, 2026; China Foreign Ministry briefing, September 14, 2026; SemiAnalysis Vera Rubin benchmark analysis; Axios Anthropic IPO reporting, September 14, 2026.