Issue #18 - April 17, 2026
- Online youth safety regulation is accelerating in the EU, boosted by member-state pressure and recent US court decisions - Mistral opens up on the AI policy framework the EU will need to succeed - A16Z suggests an impact assessment model for AI regulations to avoid overburdening "Little Tech"
Sébastien Louradour
4/22/20264 min read
Online youth safety regulation is accelerating in the EU, boosted by member-state pressure and recent US court decisions
Under pressure by many of its member states such as France, Spain, Denmark, the Netherlands and Greece — the EC has come up with the unexpected declaration that the DSA requires platforms to verify the age of users. This statement jointly made by Ursula Von der Leyen and Henna Virkkunen came with the presentation of an age verification app designed by the EC. In an op-ed in the FT, Pedro Sánchez announced that the Spanish Parliament will soon vote on a youth ban, and that a coalition of 12 member states has been leading a coordinated push at the EU level; France has already engaged in a legislative process for a ban. But to be safely enforced, privacy-enhanced solutions for age verification will need to be rolled out first. The solution presented by the EC has already been criticised for being easy to hack.
A few weeks ago, Apple announced it would roll out age verification on devices in the UK, a significant win for youth safety advocates, since Apple was not obliged by the Online Safety Act to verify the age of its iPhone users. With this new feature, Apple will be able to determine whether users are over 18 and restrict access to adult content accordingly. The UK government has since signalled it wants to go further: in a Downing Street meeting this week, Prime Minister Starmer and Tech Secretary Kendall urged social media platforms to adopt the same level of age verification already required of adult content sites. The government is currently consulting on whether to introduce an Australia-style ban for under-16s, following a House of Lords proposal — though Labour MPs voted down a Lords amendment that would have imposed one directly. The debate over how far the UK is willing to go on youth safety is far from settled.
The US has on its side landed a historic court decision on whether social media companies are responsible for the mental health of young people on digital platforms. Rather than ruling on Section 230 of the Online Decency Act, the Los Angeles court found that Meta (Instagram) and Google (YouTube) designed their products with addictive features such as infinite scroll, autoplay, and algorithmic recommendations. This judgment could represent a precedent that paves the way for hundreds of similar lawsuits.
Mistral opens up on the AI policy framework the EU will need to succeed
Mistral published this week a white paper calling for an ambitious European AI playbook. This publication is another step confirming that Mistral not only has strong ambitions as a leading AI lab but is also ready to actively participate in and drive the political debate on AI and technology. Among its recommendations, which can be read as a how-to guide to the Draghi and Letta reports, Mistral suggests creating an "AI Blue Card" — a fast-track visa specifically designed to attract researchers to Europe. Regarding the EU single market, the company calls for a simplification of the regulatory framework, which includes "eliminating or reducing" overlaps between the GDPR, the AI Act and other pieces of regulation, as well as a centralised EU AI compliance portal that would "automate compliance checks across the AI Act and GDPR". When it comes to favouring the supply of AI, Mistral supports mechanisms that would promote the adoption of EU technology over US and Chinese alternatives, and considers that EU institutions should lead by example in the adoption of such technologies. Regarding public procurement, the company advocates for establishing a targeted European preference. On data centres, Mistral considers this a key aspect of Europe's AI sovereignty, and calls for a significant effort at EU level to accelerate the deployment of a new generation of "ultra-dense AI data centres" to keep up with growing AI needs, and to allocate the decarbonised energy required to sustain them. On copyright, the company maintains the levy proposition put forward in the FT op-ed. This policy paper — the incarnation of what a European startup would need to keep competing with AI giants such as OpenAI and Anthropic — shows that the road for Europe to fully embrace the innovation journey is still long. The recent decision of some EU companies to de-risk their reliance on US AI technology has certainly represented an awaited lifeline to Mistral, which remains the most prominent EU-based frontier AI lab, but EU policies will be necessary too. For Mistral, this means two things: (i) having the EU pursue a sovereignty-focused agenda, with the risk of provoking additional friction with the US administration, and (ii) simplifying and harmonising the EU single market further — an already slow and painful journey that will require more than Mistral's voice to succeed.
A16Z suggests an impact assessment model for AI regulations to avoid overburdening "Little Tech"
On the other side of the Atlantic, another type of policy battle is unfolding over the regulation of AI. A16Z, a leading Silicon Valley-based VC whose founders Marc Andreessen and Ben Horowitz have become supporters of President Trump, has published a policy paper calling for cost-benefit analysis to be included in AI state regulations. The paper argues that most current AI regulation is not led by evidence or data. It provides a framework for evidencing the costs of regulations on businesses, arguing that while big tech can easily absorb such costs, it represents a far more delicate exercise for "Little Tech". To address this, the paper calls on policymakers to include in state bills clauses that would help evidence burden estimates, benefit evidence, and an alternatives analysis ("less-burdensome options considered and reasons they were rejected"). According to the VC, AI state laws should also mandate periodic review, with reports at six- to twelve-month intervals on whether projected costs and benefits have materialised. This ex-post evaluation would be led by a federal agency in partnership with academia and think-tanks. A voluntary cost disclosure mechanism would also be put in place so regulated businesses could report on the costs of such regulations. In addition to these legislative measures, the paper also considers other means, such as regulatory sandboxes to test regulations in limited environments and run impact assessments before fully expanding the legislation. A16Z and other US tech companies have opposed AI regulation at state level, initially pushed by Democratic-led legislatures, but a growing unease over AI risks has emerged among Republicans too — and it is likely the question of whether AI should be further regulated will feature prominently in the debates of the upcoming midterm elections. The question of simplification of tech regulation has also emerged in the EU debate. The Digital Omnibus and the Digital Fitness Check are policy files aimed at streamlining existing regulatory frameworks, but neither has introduced a comparable impact assessment framework to systematically evidence the regulatory costs borne by smaller businesses.