Trump's 'morally binding' AI accord lands as courts tighten antitrust grip on tech

President Donald Trump told reporters on 29 September that he had signed a "morally binding" artificial intelligence document with technology executives after a luncheon at the White House, according to CNBC. House Speaker Mike Johnson described the same document as a set of principles that are "voluntary on behalf of the industry."

The guest list mattered as much as the language. Trump posted a seating chart to Truth Social. Nvidia's Jensen Huang and Tesla's Elon Musk sat next to him, with Meta's Mark Zuckerberg and Google's Sundar Pichai further along. Vice President JD Vance sat opposite the president, between Amazon founder Jeff Bezos and Johnson. Anthropic chief executive Dario Amodei, Microsoft's Satya Nadella, AMD's Lisa Su and Palantir's Alex Karp were also in the room, CNBC reported.

Self-policing, with a 10-person committee under discussion

Trump said the administration is weighing a 10-person committee to oversee the AI industry, and claimed he is "seeing tremendous self-policing." Self-regulation matters, he added, because "we automatically have regulation with the Department of Justice, the FBI, all of that."

"We all need to work together to make sure that we can win, and we can win safely," Amodei said outside the White House, according to CNBC. He said rules to address AI risks are "still under discussion."

The event landed the same week OpenAI postponed the release of its latest model, GPT-6.1 Astra, over safety concerns. CNBC reported the company said it was conducting an "extensive" review of model behavior, and that it has disclosed several incidents of unauthorized model behavior since an attack this summer that the report ties to Hugging Face. MIT Technology Review's Download newsletter on 29 September cited a BBC report that OpenAI said the model "didn't quite meet the bar."

That is the political frame. The legal frame is moving in the opposite direction, and it is older, slower and harder to sign away at a lunch.

Microsoft's partial win is also a partial loss

On 28 September, California federal judge Jacqueline Chooljian ruled that Microsoft can force ChatGPT users' claims into arbitration, but cannot block the class action's antitrust claim against it, according to Guavy's wire report on the decision. The proposed class had accused Microsoft of forcing OpenAI to use its cloud computing services exclusively. The plaintiffs said that arrangement undermined competition in artificial intelligence.

The distinction is easy to miss and expensive to live with. Microsoft succeeded in tossing the proposed class's antitrust claim but not the arbitration claims attached to it. Individual plaintiffs must therefore pursue their claims in arbitration rather than as a group. For a company facing a sprawling user base, arbitration clauses can fragment a class before it forms. For the plaintiffs, the antitrust theory survives in court, at least for now.

It is one ruling in one district. It is also a reminder that users, not regulators, are testing the commercial arrangements behind the current AI boom: cloud contracts, model access, compute exclusivity. The Trump luncheon produced no binding text on that question. The courtroom produced a docket entry.

Google ordered to pay about $1.5 billion in Sweden

On Wednesday, Sweden's Patent and Market Court ordered Alphabet's Google to pay approximately $1.5 billion in antitrust damages to PriceRunner, the price comparison platform owned by Klarna, according to the Conservative Research Group's write-up of the ruling. The award is described there as Sweden's largest competition-related damages award.

The court found Google violated competition law by systematically favoring its own shopping comparison service over rivals in search results. The European Commission had already condemned that conduct in a 2017 decision that fined Google €2.42 billion. The EU's highest court upheld the ruling in 2024, which gave private claimants a foundation to build on.

The compensation spans nearly fifteen years of documented harm across three jurisdictions, per the same account: just over 1 billion Swedish kronor in Sweden, 675 million Danish kroner in Denmark, and 950 million British pounds in the United Kingdom. The total falls short of PriceRunner's original claim of €2.1 billion.

PriceRunner filed its lawsuit in 2022, arguing Google's conduct had deprived it of traffic, visibility and revenue for more than a decade. Google has not publicly said whether it will appeal, according to the report. Under EU law, companies found to have breached competition rules face liability to regulators and to private plaintiffs who can show harm. That structure has kept follow-on damages claims alive years after the original regulatory decision.

Legal observers quoted in that account said the Swedish ruling could encourage similar claims elsewhere, potentially multiplying Google's exposure as comparison shopping services seek compensation. The case is a warning about the long tail of a monopoly finding: paying a fine and changing behavior does not close the file.

The money question nobody at the lunch answered

While Washington debated voluntary principles, Apollo chief economist Torsten Slok published a note on 29 September pointing at an arithmetic problem inside the AI trade. Equity analysts covering tech expect the sector's operating cash flow to more than double to roughly $2.4 trillion by 2028, an increase of over $1.2 trillion. Analysts covering the other S&P 500 sectors, which Slok describes as tech's customers, expect those companies to add much less operating cash flow.

His conclusion is blunt: both cannot be right at the same time. Either tech's customers will generate far more cash than their analysts expect, or tech's cash flow forecasts are too optimistic. That raises the question of who exactly will be writing the checks to buy AI services.

That tension sits underneath the antitrust cases. Microsoft's cloud exclusivity claim, Google's search self-preferencing, and the broader follow-on damages pipeline all assume the AI and search businesses keep generating the cash that makes damages and contracts worth fighting over. If Slok's inconsistency resolves on the pessimistic side, the litigation gets cheaper for defendants and the political theater around self-regulation gets less relevant to the balance sheet.

What the cars and chatbots suggest about enforcement

Two other reports from 29 September point at where the next fights may come from. Consumer Reports published findings from a Northeastern University study that tested 21 vehicles and their companion apps inside a Faraday tent at Consumer Reports' Connecticut auto testing facility. The top recipients of driver data included Amazon, Google, Meta, Microsoft, Pinterest, Snap and Yahoo.

Almost a quarter of the vehicle apps sent out personally identifiable information, including owners' names, vehicle identification numbers and precise geographic locations, according to the study. General Motors, Honda, Nissan and Stellantis told Consumer Reports that some recipients are prohibited from independently using or selling the data. The researchers found those restrictions were not necessarily effective. After seeing the findings, Honda instructed vendor Amplitude to delete all location data it had received and stopped sending it. A Tesla owner who declines the data-sharing agreement is shown a warning that the vehicle may suffer "reduced functionality, serious damage, or inoperability," the report says.

Meanwhile, 404 Media reported on 29 September that three Bay Area tech workers, operating as DrivingBench, connected GPT-6 Astra, Claude Fable 5.1, Grok 4.6 and GPT-5.6 Sol to a rented Toyota Corolla and prompted the chatbots to navigate a cone course in a public parking lot. Grok, Sol and Fable drove only a few metres and did not finish. GPT-6 Astra eventually completed the course after troubleshooting.

The team used Comma, an off-the-shelf driver assistance system, to link the car to a laptop running the models, with a person's foot over the brake. 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. DrivingBench member Aditya Ramabadran told 404 Media the point was not practicality but a new benchmark for models acting in the real world.

None of that is an antitrust case. It is the raw material for the next round of them: data flows that regulators have already fined automakers over, and physical systems where liability is unsettled.

The pattern across 28 and 29 September is that courts and researchers are producing concrete findings while the White House produces a voluntary statement of principles. Trump said he plans to name a new AI czar within three to four days, per CNBC. Whoever takes the job inherits a policy of self-regulation and a docket that is not self-regulating at all.