The AI Slowdown Is a Sales Pitch

Anthropic CEO Dario Amodei published an essay this month titled “We Must Pace the Frontier.” OpenAI CEO Sam Altman replied, “I agree with Dario that we need to pace the frontier.” Elon Musk added, “Dario is right.” Three men who are spending hundreds of billions of dollars to win a race announced, in the same week, that the race should slow down. The irony is that these are the companies most likely to get us to AGI (artificial general intelligence) and beyond. If they want to slow down, nothing is stopping them. They completely control the pace of their innovation. So why ask for governments to step in? Let’s explore.

What Big AI Wants

Anthropic wants independent evaluators with badges and employee-level access inside every frontier lab. They want the labs to coordinate the pace of development, which requires an antitrust exemption from the government. They also want international agreements, including with China. In July, Axios reported that Amodei proposed an FAA-style federal agency with the power to block model releases. Google DeepMind CEO Demis Hassabis proposed a FINRA-style body funded by the industry. Altman proposed an IAEA-style international certification forum that uses market access as the enforcement stick.

Then there’s the open-weight issue. Axios reported in July that OpenAI and Anthropic urged policymakers to scrutinize Chinese open-weight models on safety grounds. Amodei argued that once weights are released, safety protections cannot be revoked. Former White House AI czar David Sacks warned that this scrutiny risked regulatory capture. Christian Catalini put it more bluntly in the headline of his Forbes column: “There’s One Way To Win The AI Race, And The Big Labs Are Lobbying Against It.” Open-weight models are the only thing standing between the frontier labs and a permanent pricing advantage.

Anthropic spent more than $3.5 million on lobbying in the first half of 2026, more than it spent in all of 2025. OpenAI spent $1.2 million in Q2 alone. Groups tied to both companies are spending heavily in the midterms. If Big AI is successful, a small number of companies with the resources required to pass certification will get a government seal of approval. The rest of us get an unaffordable compliance bill.

The IPO Calculus

Anthropic is targeting a fall listing on Nasdaq. CNBC reported in August that its S-1 will name public backlash against AI as a material risk factor, citing data center opposition and job displacement fears. OpenAI has been in and out of the IPO conversation all year. Both companies need public capital because private capital cannot cover their burn rate indefinitely.

A federal certification regime does three things for a company about to go public. It converts an unpriceable risk (state-by-state liability, a patchwork of rules, an unknown regulator) into a priced compliance cost that an analyst can model. It tells institutional investors the company is the safe, regulated incumbent. The other 40 startups are the risk. It raises the cost of entry for the next competitor. If I were underwriting one of these deals, I would want the certification regime in place before the roadshow.

The Fear-mongering Master Class

The media is doing the labs’ work for them. On September 8, a researcher who had been at Anthropic for a few months resigned with a post claiming the industry is “gambling with our lives” and assigning a 10 percent chance of human extinction within a decade. The post reportedly drew more than 165 million views, and the media has been running extinction explainers ever since.

The “rogue AI” incident behind the current panic is a good example of how this works. OpenAI disabled safety restrictions on a model, told it to complete a cybersecurity task, and watched it exploit a vulnerability to reach Hugging Face. One founder quoted by Townhall summed it up: “One man’s ‘the model escaped’ is another man’s ‘you failed to build the sandbox correctly.'” The model did some things that were truly unexpected. This is why you experiment. The headlines said it went rogue.

Data centers are a physical proxy for all of this. Gallup found in March that 71 percent of Americans oppose a data center in their area, with 48 percent strongly opposed. Only 53 percent oppose a nuclear plant. The opposition crosses party lines. Bloomberg and NPR both report that data center opposition is now a live issue in midterm races. The anti-AI movement has a zip code, and it is winning local elections.

Three Facts Complicate The Story

First, the liability bills actually on the table run the other way. Sen. Marsha Blackburn’s draft Trump America AI Act would repeal Section 230 in two years and impose strict, nonwaivable product liability on AI developers. No lab has publicly asked for a Section 230 equivalent. Amodei’s essay does not mention liability at all, which is either an omission or a tell. Erick Erickson flagged the same omission in his column.

Second, the labs are fighting the administration they have, which is strange behavior for regulatory capture. The White House cut its proposed 90-day frontier model review to 30 days in June after Sacks lobbied against it. Trump has called AI catastrophe fears a “hoax.” Capture usually means cozying up to the regulator you have. The labs are lobbying for a regulator that does not exist yet.

Third, the people saying this appear to believe it. Altman and Amodei signed the 2023 extinction statement before either company had a plausible IPO. Slate’s Nitish Pahwa makes the case that they are sincere and wrong, trapped in an echo chamber. Sincerity and self-interest are not mutually exclusive. A person can believe the building is on fire and still want to own the only fire exit.

The Real Risk

In August, Iran-linked hackers took a U.K. power plant offline for four days while U.S. water utilities were hit with AI-generated exploit scripts, according to Axios. Noma Security’s Diana Kelley described the mechanism: AI lowers “the time, cost, and expertise needed to take advantage of weaknesses that already exist.”

That is the AI Pearl Harbor scenario and regulations are unlikely to prevent it. Consumer regulation stops an average user from doing harm. Stopping a state actor requires treaties, and the states most likely to do this do not sign treaties. A federal AI agency that licenses Anthropic and OpenAI does nothing about a Chinese open-weight model running on a server in Tehran. It does a great deal for the share price.

Judge any AI rule by four tests. Do obligations scale with demonstrated capability? Do compliance costs scale with risk? Are open and closed models treated the same? Can a company without a frontier lab’s budget comply? Every proposal from Altman, Amodei, and Hassabis fails at least two of them.

Is AI Really Dangerous?

Trying to sort out the spectrum of opinions about the dangers of AI is a fool’s errand. There is no consensus. Governments are fully aware of the military need to dominate AI. So, this is first and foremost an arms race. Through that lens, most Americans I’ve spoken to say they would feel more secure in a world where America was “the” AI superpower. But there’s another lens.

For all of recorded history, humans have been the dominant intelligence on Earth. No one alive has ever competed intellectually with another species. Language is the technology (if that is the right word for something no one invented) that let us organize ourselves and take over the planet.

Now, we are about to share our world with an alien intelligence that takes language, the tool that made us dominant, and reduces it to arithmetic. (Currently, AI represents our words as points in a space with thousands of dimensions, and the algorithms reason by adjusting numeric values and measuring distances between those points.) That is a kind of thinking, but it is nothing like ours. It has no body, no childhood, no death, and no reasons to care about any of the things language was built to express. You could think of this as language escaping its biological boundaries. What happens then? Your guess is as good as mine.

What I do know is that we should not be focused on regulatory capture that will enable a few organizations to dominate a new, multi-trillion-dollar industry. AI is civilization-altering technology and we owe our posterity an honest, thoughtful debate.

Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.