Today, a Bloomberg journalist wrote about how the EU already has some oversight rules and regulations for AI (I pasted the full text below because reading the article on the Web requires a login and the story was also published in an email newsletter that is explicitly allowed to be forwarded).
EU regulation has long been the setup to a punchline. Like, how many faceless bureaucrats does it take to calculate the optimal curve for a banana?
Germany, where I’m from, is often regarded as a byword for bureaucracy. Regulations carry the stigma of stifling innovation and making it difficult for startups to take off — one of the European Union’s perennial problems, especially compared with the US.
Tech chief executive officers like SpaceX’s Elon Musk have underlined that narrative. “In the EU, for example, we find that the regulation level is extraordinarily high, and things are generally default illegal, and this inhibits progress with new technologies,” he said at the G20 Innovation Summit in early September.
When I arrived in San Francisco this summer, I expected this idea to be amplified. In the city where self-driving taxis roam the streets and you’d have a hard time meeting someone not working at an AI startup, surely innovation without constraints would be the prevailing ideal.
Instead, the talk now is all about regulation — at least a version of it that fits in with Silicon Valley’s foundational ideas. Even Musk backed Anthropic CEO Dario Amodei’s call to “slow the pace” of artificial intelligence development.With the biggest AI companies in recent days making louder calls for stricter laws, Europe could lead the way, or at least offer a starting point.
Silicon Valley’s long-standing philosophy has been to ask forgiveness, not permission. But recently, that has led to a wide array of problems, some very expensive for the companies, like Meta’s legal settlement running in the billions of dollars for the effects of social media on younger users. And the high-profile AI security breaches, which some have said are harbingers of intelligence that could potentially threaten humanity, also have AI employees and CEOs disturbed.
Politico EU described the current mood in Brussels as a “splash of Schadenfreude,” a German term for taking pleasure in someone’s misfortune. After a former Anthropic employee’s warning that AI models could kill us all by the end of the decade, European policymakers were quick to point out that the bloc already has legislation in place to prevent that: the EU AI Act.
Under the law that took effect in August, companies like OpenAI and Anthropic must take into account and counter what are called systemic risks. These include the risk of losing control of a model.
“If it gets bad or worse, we can definitely evaluate models. We can require risk mitigation measures, and in extreme cases and where necessary, we can also restrict, withdraw, or even recall AI models,” EU spokesman Thomas Regnier told AFP last week.
Now, US companies seem to be warming to the idea that a more cautious, regulated approach could lead to more predictable outcomes.
Joanna Bryson, professor of ethics and technology at Berlin’s Hertie School, advised the EU on its AI Act. She pushes back on the idea that regulation harms companies.
“The control problem becomes simpler when you externalize parts,” Bryson told me. “Of course, there can be bad external control, but it is much easier to keep safety processes intact when you don’t have to defend them internally.”
While advising a big tech company, Bryson was told how much time they spent monitoring other players in the sector, making sure they don’t misbehave. “I said: ‘This is what the government is for.’ Mature sectors have figured out how to have oversight done outside, because it is an insane amount of effort,” Bryson said. “It is a chance to throw the hard questions over the wall and let the government figure it out.”
About the US companies calling for regulations, Bryson tells Bloomberg: “There’s no reason for them to write something that’s very different from what the AI Act is.”
There’s another part of the EU legislation that seems to have left its mark on American AI companies. In his essay published last Saturday, Anthropic CEO Dario Amodei stated that the company committed to embed external evaluators such as METR – a nonprofit research organization based in California and the same organization aiding the EU in its oversight. Amodei calls on the government to require other frontier companies to do the same.
With its invisible watermark on all of Claude’s text and file output, Anthropic had already decided to comply with the EU AI Act’s transparency rules. The watermark was applied globally.
It’s too early to tell if the current mood of prioritizing safety and risk mitigation will last, but the EU’s standards could have even more ripple effects in the months ahead.
This week I attended the Dreamforce conference in San Francisco, where, in his opening keynote, Salesforce CEO Marc Benioff talked about spending the past two months in Europe. “I didn’t see a single autonomous vehicle. That was kind of a metaphor for me. I didn’t see the same kind of transformation.”
Later, he was joined onstage by OpenAI CEO Sam Altman. “When you look back at what’s happened with social media, what lessons should we take about this idea that companies should be held responsible for their actions?” Benioff asked.
Maybe he should have taken a closer look at Europe.
https://www.bloomberg.com/news/newsletters/2026-09-17/eu-s-slow-regulation-is-finding-fans-in-ai-age
And an opinion, about the concern that the AI companies currently in the lead want to use regulation to lock out competition (looks like the Amodei essay has unleashed a lot of pent-up thoughts) :
In light of AI founders like Sam Altman and Dario Amodei calling for a “pacing” of frontier technological development and the public reaction to the recent AI whistleblowers, there’s an emerging debate over the efficacy and desirability of regulation. The various proposals and calls for regulating AI have been met by insistences that regulation might lead to “regulatory capture.”
Most of these arguments have been made along the classic line that sticking regulatory bodies on the industry not only shields the current front runners from market discipline but also builds a moat that new entrants can’t traverse. But whichever particular concern is raised, these calls share a common outlook: Regulating an industry leads to regulatory capture, we’re told, because regulation is responsible for a reduction in competition, more bureaucracy, and a hit to economic growth, all of which hurts the public interest.
The historic evidence for this claim is pretty thin. And that’s because to make the claim that the regulators have been captured by the industry under regulation requires more than just observing the industry benefiting — often this is the point of regulation. It requires a demonstration that the regulators have been led astray from carrying out their public responsibilities. As the political scientist Daniel Carpenter has concluded in his assessment of the regulatory capture literature, “The process by which policy and institutions are directed away from the public interest and toward a special interest is crucial and often missing from analysis.”
It’s clear why regulation could be desirable from the perspectives of the big AI labs, who want to ward off smaller competitors. It’s also clear why they would prefer to self-regulate their “pace,” despite the imperatives of competition. But business intent cannot be the sole foundation for regulatory capture, regardless of how strong or restrictive the regulation is in implementation. To prove capture, it has to be contrasted with a defined sense of the public interest, another concept which has been pulled in many directions. Is the public interest a specific government mandate? Is it whatever improves business prosperity? Or is it the majority of Americans or even the majority of human beings?
However defined, it’s evident that regulation for the public interest has not impeded capitalism, and realistically has always been an inherent part of economic development. While it’s true that regulation could result in the current industry leaders feeling some institutional protection, as other tech giants before them have experienced, this has never stopped the tectonic shifts of creative destruction. This occurs both within an industry, as fixed capital becomes a double-edge sword along the lines of the classic The Innovator’s Dilemma , and across industries, as firms who utilize AI economize on their inputs or make other cost-saving substitutions. Regulation does not mean the end to competition, but it does mean a shift in who has a say over the social impacts of private investment decisions.
There is also historical precedent for an industry’s insiders calling for regulation. In the early 20th century, the business magnate Samuel Insull also urged the government to regulate the burgeoning electric utilities industry when his ComEd empire was expanding rapidly across the Midwest but also struggling to turn a profit as the energy market was maturing. The electrical grid was successfully regulated mostly as a public utility for about three decades until the 1970s wave of electricity deregulation.
It could instead be argued that the real regulatory capture occurring right now is the deferral by the federal government to impose some checks and balances on the industry, given that the public is becoming quite hostile to AI.
This current capture occurs in two ways. There are huge sums of money flowing into congressional races to oppose politicians calling for AI regulation — which is textbook regulatory capture. And then there is the pressure of keeping economic growth up, weighing on both the governing political party along with a fair number of members from the opposition. More specifically, it’s the capture of this particular growth model built around AI technology, which is of course ephemeral and certainly changeable, but also currently fueling much of our retirement accounts, even if the employment and income effects are fairly restricted. Both of these means of capture have led to an unresponsiveness of the current administration to the majority’s feelings on technological development.
It’s not at all obvious that the AI industry as it’s currently constituted, with the number of players or the flow of funding rushing in or the purported importance of the technology as a general purpose input to all other industries, will remain this way in the near future. But given the emerging mass concern over the quality, utility, and safety of the industry’s product, it seems likely that regulation will play an increasing role in the pace of AI development, much like it was for electricity.
https://www.bloomberg.com/news/newsletters/2026-09-17/regulatory-capture-isn-t-what-you-think-it-is