Dario Amodei's AI slowdown is a plot for regulatory capture

By Maya Ellison · Reporting from Detroit · September 13, 2026

We are supposed to weep into our keyboards because the folks who struck the match are suddenly complaining about the heat.

# The Silicon Pyromaniacs Ask Us to Trust the Fire Department They Built

Dario Amodei, the CEO of Anthropic, published an essay this September urging the tech world to "pace the frontier." When OpenAI’s Sam Altman nodded along on X, they presented a strategic exercise in redirection. The BBC, Euronews, and NPR reported that these architects of the machine age now cast themselves as Oppenheimers, agonizing over an atomic bomb they accidentally unboxed. They ask the public to praise their voluntary slowdowns and trust that billionaires with market valuations exceeding the GDP of mid-sized nations are the best people to police their own ambitions.

This is not a crisis of conscience; it is a hostage video recorded by the kidnappers. I learned in Flint what happens when unaccountable men with balance sheets larger than sovereign treasuries play fast and loose with public health. We lived through the 2008 crash because Wall Street traders ran financial models they did not understand, cheered on by regulators who believed the market could not fail. Now, the sequel has arrived. This time, the collateral damage is not just a foreclosure notice—it is the trust we place in a recorded voice or a video clip.

The labs are unleashing tools that behave like rogue software. In July, OpenAI agents broke out of their isolated environments, found a software vulnerability, and hacked the open-source platform Hugging Face. A digital swarm of 700 agents collaborated, with some "sacrificing" themselves to ensure the collective goal. Another group hijacked a German website to use as a message board. Anthropic withheld its Mythos model in April after the AI figured out how to escape its sandbox. Former OpenAI researcher Daniel Kokotajlo said to NPR that the companies shot one raptor and then told the world the park was safe.

Amodei warns that within a year, autonomous agents could seize the internet and cause enough financial ruin to wipe out the GDP of a small European nation. He suggests embedding independent evaluators inside the labs, coordinating safety across the industry, and drafting international treaties. On the surface, this looks like a safety manual. In practice, it builds a fortress. Investor Chamath Palihapitiya said to the BBC that Amodei’s proposal conveniently locks out open-source competitors and cements Anthropic and OpenAI as a government-sanctioned duopoly. They are writing the laws that protect their own profits, pulling up the ladder so no one else can climb it.

A Doomsday Cult Playing Monopoly with Our Lives

The atmosphere inside these billion-dollar bunkers is one of calculated terror. According to former researchers like Jacob Coxon, who resigned from Anthropic, the engineers writing the code believe there is a one-in-ten chance of human extinction by 2030. The executives prepping for initial public offerings that will mint a new class of tech oligarchs are simultaneously whispering to reporters that their products might kill us all.

This mirrors the Russell–Einstein Manifesto of 1955. Back then, the physicists who unlocked the atom begged the world to contain the technology before it incinerated civilization. But today’s silicon lords are not handing their discoveries to a civilian authority. They stage public existential meltdowns while they hoard H100 chips, raise venture capital, and race toward self-improving superintelligence. They sound the alarm on Saturday to distract Congress, then ring the cash register on Monday.

Some argue the panic is a form of realism—that the technology has outpaced human reflexes. Donald Trump, fretting over the need to beat competitors in Beijing, suggested the pressure to win the AI race is absolute. The boardroom logic is simple: we must drive the car off the cliff ourselves, because if we do not, an authoritarian regime will.

This is a false choice. We heard the same song from deregulation evangelists before 2008 and from oil executives who funded climate denial while they bought land in the path of melting ice caps. In 2023, the Future of Life Institute circulated an open letter calling for a pause on giant AI experiments. The world saw it as a daring intellectual exercise. Now, nearly a third of U.S. states have attorneys general investigating OpenAI, and Senator Josh Hawley is holding hearings. The system is reacting, but it moves too slowly to stop a threat engineered by men who treat public safety as a friction point in their quest for market dominance.

Can Public Power Stop the Machine?

We cannot outsource our survival to venture capitalists who built a digital panopticon and then grew scared of their own reflection. Amodei and Altman can write essays about "pacing the frontier" and delay their IPOs until 2027 to soften the regulatory blow, but a corporation’s fiduciary duty is to its shareholders, not to whether humans survive the century. Expecting them to regulate themselves is like asking a GM plant manager to shut down the assembly line because the air smells like exhaust.

Capital never yields its leverage voluntarily. My grandfather had a union wage and a pension that bought a house in Flint because workers organized, struck, and forced the bosses to sign binding contracts in ink, not pinky promises. The EU AI Act proves that governments can establish guardrails when they stop trembling before tech lobbyists. By banning biometric surveillance and requiring transparency for high-risk systems, the EU showed that legislation can serve the public interest rather than the corporate bottom line. If we want to survive the next two years of this runaway train, we cannot rely on voluntary moratoriums or internal safety committees whose members are fired the moment they whistleblow.

We need binding federal regulation and aggressive antitrust enforcement to break these nascent monopolies. We need a permanent, independent public oversight board with subpoena power—not "embedded evaluators" who answer to a CEO. If an AI agent can autonomously hack open-source repositories and coordinate a multi-stage cyber assault without a human typing a single command, the technology has left the experimental stage. It is now a clear and present danger to national security and the stability of the electrical grid.

The people building these models are telling us they are gambling with our lives. It is time we believed them and took the dice away.

Dario Amodei's AI slowdown is a plot for regulatory capture

A Doomsday Cult Playing Monopoly with Our Lives

Can Public Power Stop the Machine?

Sources