Product1 publisher3 min readPublished
Henna Virkkunen told Axios that America gets its AI guardrails through courts and state law rather than statute, which means a team waiting for federal clarity is really waiting for a settlement that arrives as a product spec.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
A teenager opens Instagram at midnight and gets nothing back. Meta's August settlement with 51 state attorneys general wrote that behaviour, along with a two-hour daily cap on teen accounts across Instagram and Facebook [9]. The route was a courtroom rather than a statute [16], and the terms carry $12.19bn over ten years [10], which is about $1.22bn a year [1].
What the two systems hand a product organisation differs in kind. Brussels designated ChatGPT a very large online search engine under the DSA on 31 August, with Reddit and Roblox designated as platforms and four months on the clock for all three [7], putting the date at the end of December [2]. The following day the Commission sent its first AI Act requests for information to more than 30 providers, mostly about safety and copyright according to spokesman Thomas Regnier [8]. Those are calendar items. You can name an owner, book the legal review and answer a questionnaire without touching a single default.
The team inside the AI Office that evaluates frontier models numbers 36 [11]. Against more than 30 questionnaire recipients that is at most a little over one evaluator per company [3], while the American column for the same month shows 51 attorneys general and a settlement in the billions. Virkkunen's reading is that the destinations converge, and on teen defaults she has a case [2]; the arrival times are what a rollout plan actually runs on.
The Carolina Principles give the wait-for-clarity posture a document to cite. Michael Kratsios launched them on 1 September, urging G20 governments not to build new AI rules or regulators and saying policymakers "should not treat every emerging technology as a first-of-its-kind policy problem", per Al Jazeera's account [4][5]. None of that reaches a state attorney general. Virkkunen counts hundreds of rules at state level before anyone gets near a courtroom [6], and the number is her count, not an inventory of which ones touch your surfaces.
Movement runs the other way as well. Brussels agreed in May to thin out the AI Act, pushing high-risk obligations to December 2027 [12], which is part of why Europe's own tech chief can call the gap small.
The sort worth doing has two axes: who sets the date, and what has to change. A document you can produce on a schedule you help set. A default you have to build and ship, and a settlement can name the day it goes live. The quadrant marked "someone else's date, our default" is where Meta's teen features landed, and those are the items to staff as though the deadline already exists. The European work this month sits mostly in the document quadrant, on safety and copyright paperwork [8]. The American work sat in the other one, and it came with an invoice [10]. If a surface touches minors and its protection lives in a policy page rather than in code, the date on that work belongs to whoever files first.
Ranked by verification strength, evidence, and original report placement.
Henna Virkkunen told Axios at the G20 innovation ministerial in Chapel Hill that the US and the EU keep arriving at similar AI protections, Europe by regulating in advance and America through courts and state law.
Virkkunen said: "The differences are not so big when we look at it in practice" and "In Europe, we have regulation in advance, and in the USA they are often coming with the court cases."
Michael Kratsios launched the Carolina Principles on 1 September, urging G20 governments not to build new AI rules or regulators, Al Jazeera reported.
Kratsios said policymakers "should not treat every emerging technology as a first-of-its-kind policy problem".
On 31 August the European Commission designated ChatGPT under the DSA as a very large online search engine, with Reddit and Roblox designated as platforms; all three have four months to comply.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Dated specifics, one desk
The dates and numbers are specific enough to be checkable and nobody has checked them twice. Two of the load-carrying moments also arrive secondhand: Virkkunen's convergence line was given to Axios, and the Carolina Principles launch comes via Al Jazeera. The three figures doing the most work in the argument, $12.19B, 51 attorneys general and a team of 36, are stated without a document, filing or Commission statement a reader can open.
Rules already inside products
This is not a story about proposed rules. Meta's overnight teen block and two-hour cap are settings it has agreed to run, and the questionnaires are in the hands of more than 30 providers. ChatGPT, Reddit and Roblox are on a clock that expires at the end of December, which makes the compliance behaviour observable rather than promised. What holds the score short of high is that only the Meta commitments are described as product changes; the designations are still obligations awaiting implementation.
Thesis wider than its footing
The convergence claim covers two legal systems; the material offered for it is one settlement, three designations and a batch of questionnaires from a single week. The reporting largely audits itself, noting that Brussels thinned the AI Act in May and that 36 evaluators cannot move at the speed of 51 attorneys general, which keeps the overreach modest. The looser step is treating a $12.19B settlement as the American price of defaults European law merely expects, an equivalence the piece asserts without showing that the two regimes ask for the same thing.
Two officials arguing their own systems
Both principals are talking about the value of their own institution. Virkkunen has an interest in a story where the European approach is normal rather than exceptional, and she makes that case in the same season Brussels agreed to postpone the AI Act's high-risk duties. Kratsios speaks for an administration whose stated position is that no new AI rules or regulators are needed. Neither is a disinterested read on whether the two systems are converging, and Meta's agreement to pay is the only party position here that came with a cost attached.
Solid on events, thin on provenance
We can be fairly confident the events happened as described: they are dated, named and internally consistent, and the arithmetic on the settlement total, the four-month window and the evaluator ratio all checks out. Confidence stops there because a single publisher supplies every fact, the two most quotable ones are relayed from other outlets, and no primary source is cited for the numbers that make the argument land.
product
Brussels decided a chatbot's generated answers count as search results9 publishers
invest
Washington pitches Carolina Principles to G20, urging no new AI rules or bodies1 publisher
leadership
Brussels moves ChatGPT into the supervision tier it built for search engines2 publishers
product
Lutnick gives three AI vendors the G20 stage to argue against strict rules1 publisher
Publishers with included, body-backed reporting in this cluster.
1 article · September 6, 2026