Teen ban details emerge, how LLMs view autocracies and The Lancet on misinfo

Hello and welcome to Everything in Moderation's Week in Review, your need-to-know news and analysis about platform policy, content moderation and internet regulation. It's written by me, Ben Whitelaw and supported by paid members like you.

Despite writing EiM most weeks since 2018, I still get a buzz out of seeing who joins its readership each week — and none more so than this week. A warm welcome to new subscribers from Wikimedia, the Computational Democracy Project, GIFCT, the eSafety Commissioner, Ofcom, Aibi.ai, Minalab, Coimisiún na Meán, Bumble and elsewhere.

If you're been following the newsletter for some time, there's never a better time to support independent coverage of the T&S industry. Become a member — and if you can't afford it, get your employer to stump up.

In this week's Ctrl-Alt-Speech, I sat down with Niklas Eder, User Right's co-founder and co-CEO, to read the tea leaves on Europe's plans for a harmonised social media ban and its latest Big Tech enforcement play. Out now for Patreon supporters, arriving later today for free listeners.

Here's everything in moderation from the last seven days — BW

in partnership with Safer by Thorn, a purpose-built CSAM and CSE solution

New feature! Context Labels add more information to suspected CSAM.

As part of Thorn’s continued commitment to helping trust and safety teams detect and disrupt child sexual abuse material (CSAM), we're happy to announce that Safer now includes Context Labels, a new feature that provides additional signals for suspected CSAM.

These new Context Labels give trust and safety teams greater clarity when reviewing suspected CSAM by providing three predictive signals for nudity, apparent maturity, and sexual content. This gives your team more context during moderation, so you can:

• Triage content more efficiently
• Prioritize high-risk cases
• Make more informed moderation decisions

These labels can integrate into your existing moderation workflows to help your team filter, sort, and prioritize flagged content, whether you're using Safer directly or incorporating results into your own internal review tools.

LEARN MORE

Policies

New and emerging internet policy and online speech regulation

There is finally some flesh being put on the bones of social-media age restrictions — although the UK and EU are building rather different creatures.

• UK: 16 and 17-year-olds will be unable to access social media between midnight and 6am after government research showed it was the best way to reduce usage — but it will be able to be turned off in account settings. One expert called it a “mildly annoying settings prompt with a government press release attached". The government said it wants to disable addictive features like autoplay and infitite scroll but would be no VPN ban as there was little evidence it was being used. Phew. However, there’s still no clarity on the platforms that will be in and out of scope.
• EU: The much-awaited Special Panel on Child Safety has advocated "access restrictions" for under-13s within the bloc with the potential for national measures for 13-18 year olds. The approach relies heavily on the EU-wide approach to age assurance, which you may remember had a stuttering start back in April (EiM #333). President Ursula Von der Leyen’s line — that the debate is not only about whether children can access social media, but “whether and when social media can access our children” — was inevitably repeated across the coverage.

The difference in the two approaches — the UK’s piecemeal, drip-fed ban and the EU’s considered, more warmly received approach — will not be a huge surprise to policy onlookers who saw the emergence of the Digital Services Act and Online Safety Act. But it further underscores how successive British governments have bodged tech policy. It would be a remarkable shock if Prime Minister-in-waiting Andy Burnham was any different. This week's Ctrl-Alt-Speech has more.

Also in this section...

• Let’s build a children’s public internet (The Verge)
• Commission accepts X's corrective measures to terminate breaches of the DSA (European Commission)

Products

Features, functionality and technology shaping online speech

I haven’t finished digesting the Oversight Board’s new assessment of large language models — it only was published yesterday — but its core finding is worth flagging. It finds that leading AI systems are significantly less willing to generate criticism of restrictive governments than that of democratic ones. The Board also warns that, unless frontier AI labs take human rights due diligence more seriously, widely-used LLMs risk extending illegitimate speech restrictions. 

Why it matters: The 10 models assessed — including those from Anthropic, Google, Meta, OpenAI and others — are fast becoming the upstream layer of content moderation for many intermediaries. Their policies could become inherited by major platforms but also thousands of smaller services. It means that the question “Does this platform or service moderate speech fairly?” needs expanding to the systems on which platforms increasingly depend.

Also in this section...

• Safer Context Labels: More clarity for CSAM moderation (Thorn)
• Can AI moderate social media content? Here’s why it falls short (The Hindu)

Platforms

Social networks and the application of content guidelines

I mentioned it briefly in last week’s newsletter (EiM #343) but the European Commission has found Meta in breach of the Digital Services Act for its “addictive design”, a mere 26 months after opening its investigation. Features such as autoplay and infinite scroll — yep, them again — “shift the brain into autopilot mode, contributing to unhealthy habits and compulsive use”. Although the release has no further details of how they came to that conclusion.

Product of our times: Seen alongside the EU’s finding against TikTok from February (EiM #324), it’s clear that product design is a priority for the Commission. But my sense from speaking to legal experts is that this won’t go far: not only does the DSA doesn’t explicitly define “addictive design” but it’s also a core part of the Digital Fairness Act, the flagship law arriving later this year which is product-heavy in its approach.

As such, I thought the take from Weizenbaum Institute’s Julian Morgan was a solid one; that the decision “makes more sense read as a strategic opening move in a negotiation than as the enforcement of a settled legal standard”.

Also in this section...

• Why teens deserve access to safe AI (OpenAI)
• It’s Game Time for Content Moderation (CDT)

People

Those impacting the future of online safety and moderation

Misinformation appears to have lost some of its prominence within trust and safety community. As Alice highlighted in this week’s T&S Insider, just two sessions are explicitly about it, a drastic decrease from just a few years ago. The problem is clearly not solved but, due to company and funding priorities, it’s no longer attracts the same attention.

But that doesn’t mean it’s not a concern outside the platform policy space. The Lancet, one of the world’s leading medical journals, has announced a new Commission to look into misinformation, health and human security. Such commissions tend to have between 20-25 experts from diverse fields and produce a weighty report with policy recommendations that governments and non-profits can take forward.

Professor Heidi Larson, the director of the Vaccine Confidence Project at a London School of Hygiene & Tropical Medicine (LSHTM), will co-lead the commission and, based on the scope set for herself, has a big job on her hands. While major US-based platforms have decided that mis- and disinfo are less of a priority, the need for practicial interventions remains as acute as ever.

Posts of note

Handpicked posts that caught my eye this week

• “For the past two years, my lab has been exploring how to create online spaces designed for a basic human activity: conflict. See what we've learned on August 12, when we release our new zine, Conflict Systems, at the Metagov Seminar” - University of Colorado’s Nathan Schneider knows I love a zine. 
• “For over a decade, I worked on child safety and public policy inside some of the world's biggest technology companies, helping think about how we make digital spaces safer for young people.” - Former Google and Netflix public affairs expert Catherine Williams on working with schools and youth organisations.
• “Now, we could not be more excited that, with the tremendous support of the Templeton World Charity Foundation we'll be able to adapt that model and host a researcher<>practitioner convening with professionals building community and connection in the Open Social Web.” - I’m a big fan of the Julia Kamin's work with the Prosocial Design Network and its new initiative only adds to that.