Anthropic Warns AI Could Control Civilization

Anthropic Warns AI Could Control Civilization, China Immediately Asks Where to Download It

American AI Safety Campaign Accidentally Becomes DeepSeek’s Most Effective Recruiting Brochure

The international campaign to convince China that artificial intelligence is extraordinarily powerful, potentially civilization-changing and perhaps capable of determining which nations dominate the twenty-first century has hit an unexpected snag. Chinese researchers keep responding, “You had us at extraordinarily powerful.”

ChinaTalk reported this month that some Chinese researchers and commentators increasingly view Anthropic not merely as another American technology company, but as a symbol of technological concentration and American attempts to restrict Chinese access to frontier AI. DeepSeek researcher Liu Shengyu wrote that competition leaves him feeling compelled to keep advancing AI even though he expects the technology eventually to surpass his own kernel-optimization work.

Thus humanity has reached an important AI milestone: engineers are now working overtime to invent the machine that eliminates the need for engineers to work overtime.

Economists traditionally call this productivity. Employees traditionally call it Tuesday.

AI Safety Becomes World’s First Advertising Campaign Built Entirely on Terror

Anthropic CEO Dario Amodei has spent years arguing simultaneously that powerful AI could produce extraordinary benefits and extraordinary dangers. In his 2024 essay “Machines of Loving Grace,” he described a future in which powerful AI could accelerate advances across health, neuroscience, economic development, governance and work.

More recently, Amodei called for slowing the rate of frontier AI development so safety systems could catch up, warning that capabilities were advancing extremely rapidly.

This presents a small marketing difficulty. Ordinarily, manufacturers advertise products by saying “now 30 percent quieter” or “gets whites whiter.” Silicon Valley has developed another method: “This machine may reorganize civilization, transform military power, replace millions of jobs and become smarter than virtually everybody you know.”

Chinese engineers responded exactly as any reasonable competitor would. “Interesting. What model number?”

Anthropic has therefore invented the first safety warning that doubles as a product demonstration. Cigarettes say SMOKING KILLS. Frontier AI apparently says MAY CONTROL THE FUTURE OF HUMAN CIVILIZATION. One discourages purchases. The other causes governments to order 80,000 GPUs.

Dario Amodei Accidentally Named DeepSeek’s AI Employee of the Month

There may be no greater motivational speaker for Chinese AI development than an American technology executive explaining exactly how valuable AI superiority could become. Amodei has repeatedly argued for restrictions on China’s access to advanced computing technology, while Chinese critics increasingly interpret such measures as an attempt to preserve American technological dominance.

Imagine trying to discourage your neighbor from digging for oil by shouting over the fence: “STOP DIGGING! THERE COULD BE TRILLIONS OF DOLLARS OF OIL DOWN THERE!” Five minutes later the neighbor owns a backhoe.

American message: “Artificial intelligence could determine national power for generations.” Chinese engineer: “Cancel dinner.” American message: “These models could become smarter than Nobel Prize winners.” Chinese engineer: “Cancel breakfast too.” American message: “We may need export controls because this technology is so strategically important.” Chinese engineer: “Somebody order another coffee machine.”

Nashville comedian Denny Rourke put it best at a recent open mic: “We spent a decade telling China not to touch the stove. Turns out the stove was self-cleaning, self-improving, and hiring.”

Everyone Opposes the AI Arms Race While Accelerating Through the School Zone

Liu’s dilemma, as described by ChinaTalk, is beautifully human. He believes AI may eventually equal or surpass the specialized kernel work he spent years mastering, but slowing his own efforts would not halt competitors. So he keeps racing.

Every participant has arrived at the same philosophical position: the race is terrible, and I intend to win it. This is six motorists passing one another at 115 mph while lowering their windows to shout “PEOPLE DRIVE TOO FAST!”

Washington worries Beijing will race ahead. Beijing worries Washington will race ahead. Anthropic worries everybody may be racing too quickly. DeepSeek worries somebody else may develop the technology first. The AI presumably watches this and concludes that human beings have already solved alignment. Everybody is perfectly aligned toward accelerating.

Engineer Devotes Career to Building the AI That Makes His Career Unnecessary

The most exquisite detail of the entire story may be Liu’s expectation that AI could soon match or surpass his own kernel optimization abilities. This is no longer automation. This is professional assisted self-replacement.

Historically, a blacksmith occasionally worried about machinery replacing him. He generally did not spend Saturday afternoon improving the machine. Modern AI engineers have developed a more entrepreneurial approach.

“Will this eliminate my job?” “Possibly.” “Excellent. I found another 11 percent performance improvement.”

LinkedIn may eventually add a new employment category: Senior Director of Making Senior Directors Unnecessary. The annual performance review should be fascinating. MANAGER: “Where do you see yourself in five years?” ENGINEER: “Economically redundant.” MANAGER: “Excellent strategic awareness.”

AI Offers Humanity Two Futures: Communism or Cyberpunk 2077

Liu’s essay contemplated two dramatically different AI futures: broad technological abundance or a world where a small number of companies control the most powerful technology. He used Cyberpunk 2077 as the dystopian reference point.

Every intermediate possibility has apparently been cancelled. No Switzerland. No moderately regulated capitalism. No Costco. No pension fund earning 7 percent annually while complaining about management fees.

Civilization now gets two options: technological abundance distributed broadly, or Jeff Bezos acquiring laser eyeballs while your refrigerator demands cryptocurrency. Economists, having spent three centuries developing nuanced theories of markets and institutions, have reportedly been reassigned to determine whether cybernetic arms are deductible.

Anthropic’s AI Somehow Becomes the Final Boss of Humanity

The name “Anthropic” derives from the idea of humanity. That branding once sounded reassuring. It suggested thoughtful people sitting around discussing human welfare. ChinaTalk describes a very different perception taking hold among parts of China’s online tech community, where commenters have portrayed Anthropic as an ideological and technological gatekeeper rather than merely an AI company.

This is a remarkable corporate journey. You establish a company built around humanity’s welfare. You hire philosophers. You create a safety team. You publish careful essays about responsible technology. Five years later somebody paints you on the side of a building holding Planet Earth over a volcano.

Open Source AI Becomes Marxism With Graphics Cards

Chinese supporters of open AI increasingly frame accessible models as protection against technological concentration, according to ChinaTalk. Liu himself argued that frontier intelligence should be broadly and cheaply available rather than controlled indefinitely by a few companies.

Karl Marx once called for workers to control the means of production. In 2026 the means of production apparently requires CUDA drivers. The revolutionary manifesto has been updated: workers of the world, unite, you have nothing to lose but your rate limits. Today’s barricades contain liquid-cooled GPU racks. The proletariat has a GitHub account. And history’s next revolution may be delayed because somebody installed the wrong version of PyTorch.

The AI Cold War 2.0 Will Be Conducted Entirely Through Blog Posts

The geopolitical structure is becoming wonderfully efficient. American analysts read Chinese government statements and perceive strategic competition. Chinese analysts read American export-control proposals and perceive containment. American researchers read Chinese criticism and perceive hostility. Chinese researchers read the American response and perceive confirmation.

Soon nobody will need intelligence agencies. Analysts will simply Substack their way through a crisis that used to require satellites. ChinaTalk argues that distrust increasingly causes both sides to interpret the other’s actions through geopolitical motives, even when real disagreements within each country complicate those narratives.

The original Cold War required submarines, bomb shelters and thousands of nuclear warheads. Cold War 2.0 requires Substack Pro.

Boy Who Cried Wolf Announces Bigger AI Wolf Available Tuesday

Chinese skepticism toward American AI safety warnings presents another fascinating problem. ChinaTalk cites Chinese online responses mocking recurring warnings that frontier models could become dangerous, while some observers interpret those warnings through the existing geopolitical rivalry.

It is the Boy Who Cried Wolf problem, except the boy owns a laboratory dedicated to building increasingly sophisticated wolves. Village: “You keep exaggerating.” Boy: “Fair criticism.” Village: “So there isn’t a wolf?” Boy: “No, there absolutely is. We just upgraded it. Wolf 4.5. Better reasoning. Improved coding. Forty percent less hallucination.” Village: “Can it eat us?” Boy: “That remains an active area of safety research.”

America Discovers Worst Possible Way to Make China Lose Interest in AI

Suppose you want children to ignore a cupboard. You can say “nothing interesting in there.” Or you can say: “Inside that cupboard is the most valuable, powerful and potentially transformative cookie ever created. Whoever possesses it may dominate all future snack distribution. Under absolutely no circumstances may you open it.”

Within four minutes, six graduate students from Tsinghua University are examining the hinges.

The strategic paradox is unavoidable. The more dramatically America describes frontier AI’s importance, the stronger the incentive for competitors to obtain it. The more Washington attempts to restrict advanced technology, the more valuable technological independence appears to Beijing. And the more Chinese policymakers describe American restrictions as containment, the easier it becomes for American officials to interpret Chinese development as confirmation that restrictions were necessary.

That is not a feedback loop. That is a feedback Ferris wheel, and everybody bought unlimited tickets.

Artificial Intelligence Successfully Recreates Curtains

After trillions of dollars in expected investment, billions of model parameters and thousands of the world’s brightest scientists, humanity has produced machines capable of translating languages, generating software, analyzing scientific data and conducting increasingly complicated reasoning.

Naturally, civilization immediately used them to reinvent humanity’s oldest geopolitical technology: suspicion. American side: “What are they doing?” “Developing AI.” “Why?” “Probably because of us.” Chinese side: “What are they doing?” “Restricting AI.” “Why?” “Probably because of us.”

And somewhere inside a datacenter, an artificial intelligence analyzes several million pages of diplomatic history before producing the only conclusion consistent with available evidence: “Have either of you considered talking?” The machine is immediately unplugged for displaying dangerous emergent behavior.

There is, beneath the comedy, one genuine point worth keeping. ChinaTalk notes that Chinese and American actors still identify some overlapping concerns, including reliability, biological and cyber misuse, autonomy and maintaining meaningful human control. China’s own 2023 Global AI Governance Initiative called for improving AI predictability and keeping the technology under human control. So the world’s two great AI competitors may actually agree on one thing: artificial intelligence should remain under human control. Unfortunately, nobody has yet demonstrated that humans remain under human control.

Sources

This is American satire: Anthropic, DeepSeek, Dario Amodei, Liu Shengyu and the GPUs are real; the conversations, motives and cybernetic ambitions attributed to them here are not. Consider it American satirical journalism doing what satirical news has always done: sprinting past the press release to get to the punchline.

For the British satirical take on the same AI panic, visit The London Prat at prat.uk.

Auf Wiedersehen, amigo!

Bohiney News