White House AI Hands-Off Deal Leaves Antitrust Questions Open

Trump told reporters after the luncheon, which House Speaker Mike Johnson also attended, that he is "seeing tremendous self-policing" and that his administration is weighing a 10-person committee to oversee the AI industry. According to CNBC, Johnson described the document as a statement of principles that are "voluntary on behalf of the industry," with the White House guiding development.

The event drew Anthropic CEO Dario Amodei, Nvidia's Jensen Huang, Tesla's Elon Musk, Meta's Mark Zuckerberg, Google's Sundar Pichai, Microsoft's Satya Nadella and Amazon founder Jeff Bezos, among others.

A seating chart posted to Trump's Truth Social account placed the president next to Huang and Musk, with Zuckerberg and Pichai alongside. Trump also said he plans to name a new AI czar within three to four days. The gathering took place as multiple AI-related bills have been introduced in both the House and Senate in recent weeks, according to CNBC. None of those proposals was mentioned in the agreement signed on 29 September, and no enforcement mechanism was described.

Voluntary principles, no enforcement

Outside the White House, Amodei said rules to address AI risks are "still under discussion." "We all need to work together to make sure that we can win, and we can win safely," he said, according to CNBC. Palantir CEO Alex Karp, speaking to CNBC ahead of the event, said "the main issue that you have and I have is we have to take responsibility for the dangers we're aware of."

AMD CEO Lisa Su said she was "very encouraged" by the event and described "a lot of optimism and a sense of responsibility" in the room. "At the end of the day, it's our responsibility to show the power of the technology as well as ensure that it's very safe," Su told reporters.

"There's a belief that there should be tremendous self-regulation, and we automatically have regulation with the Department of Justice, the FBI, all of that. But the self-regulation is very important."

That quote, from Trump to reporters after the luncheon, is the clearest statement of the administration's position. It is also the point where the antitrust thread runs into difficulty. The Department of Justice and the FBI enforce criminal and competition law, but neither agency was named as a participant in the agreement, and no new authority was created.

Trump also called multibillion-dollar data centers a "very positive thing" despite local opposition ahead of the midterm elections, and said tech leaders "want to see communities that are safe and happy." "Otherwise, they'll be forced to go overseas or other locations," he said, according to CNBC.

The money does not add up

The same week, Apollo chief economist Torsten Slok published a note arguing that sell-side forecasts for the technology sector are internally inconsistent. Analysts covering tech expect operating cash flow to more than double to roughly $2.4 trillion by 2028, an increase of over $1.2 trillion, Slok wrote on 29 September.

Analysts covering the rest of the S&P 500, which are tech's customers, expect those companies to add much less operating cash flow. "Both cannot be right at the same time," Slok wrote. "The bottom line is that either tech's customers will generate a lot more cash than their analysts expect, or tech's cash flow forecasts are too optimistic, which raises the question of who exactly will be writing all those checks to buy AI services."

The note is not an antitrust argument, but it points at the same concentration question from a different direction. If the demand for AI services has to come from companies whose own cash flow forecasts are modest, the case for the current scale of investment rests on a narrow set of buyers. That is a market-structure problem, and market-structure problems are what competition authorities are supposed to examine.

Slok's framing also undercuts the self-regulation case. A voluntary agreement among the largest labs and cloud providers does not address who pays for the compute, or what happens if the customers cannot.

What the cars say

A study from Northeastern University researchers, conducted with Consumer Reports vehicles and its Connecticut auto test center, found that data from cars and their companion apps flows to some of the largest technology companies, including Amazon, Google, Meta, Microsoft, Pinterest, Snap and Yahoo. Consumer Reports published the findings on 29 September; the study is due to be published this week.

Almost a quarter of the vehicle apps tested were sending personally identifiable information, including owners' names, vehicle identification numbers and precise geographic locations. "Whether or not consumers are aware, big tech companies are all over the vehicles that we drive," said Nicole Zagson, a doctoral candidate in cybersecurity at Northeastern and a study co-author.

Another co-author, Sarah Elizabeth Gillespie, said: "It does not appear that a customer can buy a new car that does not track you." The researchers tested 21 vehicles from model years 2022 through 2025 inside a Faraday tent that blocks electromagnetic signals.

Several automakers told Consumer Reports that some links in their connected apps open outside webpages where third parties can embed pixels and cookies. General Motors, Honda, Nissan and Stellantis said some recipients were prohibited from independently using or selling the data. After the researchers showed their findings to Honda, the company instructed a vendor, Amplitude, to delete location data it had received and stopped sending it.

Tesla owners who decline the company's data-sharing agreement are shown a warning that it "may result in your vehicle suffering from reduced functionality, serious damage, or inoperability," according to Consumer Reports. That is the consent problem in one sentence: the alternative to sharing is not a functioning car.

An industry that says it is already dead

On the same day, a post on the Monroe Lab site titled "The Tech Industry: An Autopsy" argued that the sector has destroyed itself in its quest to destroy labor. "In its quest to destroy labor, it is, supreme irony, destroying itself," the post says. It is a polemic rather than reporting, and it offers no data, but it captures a mood that the White House luncheon did not dispel.

A book published this week by Haymarket Books, "Against Tech Oligarchy" by JS Tan and Clarissa Redwine, chronicles the tech worker movement of the past decade. In an interview with the Bay Area Current, the authors describe the "Californian Ideology" as a fusion of "the utopian visions of both the New Left and the conservative right, celebrating market power and individual empowerment while refusing to critically examine the contradictions between them."

The book traces campaigns against Project Maven at Google, walkouts at Riot Games and organizing at Kickstarter, Amazon and the New York Times. Its argument is that worker pressure, not regulation, produced the sector's few accountability moments.

That reading sits awkwardly next to the 29 September agreement. If the past decade of tech accountability came from employees and public campaigns, a voluntary document signed by chief executives and enforced by no one is a thin substitute.

Frontier models on a cone course

Also on 29 September, 404 Media reported that three Bay Area tech workers, who call themselves DrivingBench, hooked up GPT-6 Astra, Claude Fable 5.1, Grok 4.6 and GPT-5.6 Sol to a rented Toyota Corolla and let the models steer through a parking-lot cone course. Grok, Sol and Fable drove only a few meters; GPT-6 Astra eventually completed the course.

Aditya Ramabadran, Tobias Gessler and Simon Mahns met at Axiom Math, an AI math startup, and used an open-source Comma kit to connect the car to a laptop running the models. "The point wasn't to show that it's practical for you to plug ChatGPT into your car and have it drive you places," Ramabadran told 404 Media. The National Highway Traffic Safety Administration announced an investigation into Comma last week after two crashes involving the system killed three people, 404 Media reported.

This is a stunt, and the team says so. But it is also a benchmark claim: that general-purpose models can now act in the physical world with no prior training data. If that claim holds, it strengthens the case for the safety review that OpenAI itself invoked when it postponed GPT-6.1 Astra over safety concerns, a delay CNBC reported the same week.

MIT Technology Review's The Download noted on 29 September that Anthropic's new molecular biology lab claimed its first discovery, a pattern around an enzyme, and that the claim angered biologists. One researcher said his team had already found the same pattern. "It's a reminder that what's novel for AI may be routine, unsurprising or simply not that consequential to a biologist," MIT Technology Review wrote.

Put the two together and the pattern is clear. The companies signing voluntary principles at the White House are the same companies making contested claims about what their models can do. Neither the agreement nor the bills in Congress set out who checks those claims, or what happens when they are wrong. The antitrust question, about who controls the compute and the data and the customers, is not answered by a document that is binding only in a moral sense.