OpenAI joins the club, FTC builds its case and 153m reasons to worry

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.

Some things don't surprise me anymore; teen products are launched without substantive evidence, Andrew Ferguson throws his weight around, identity data is leaked. Plus ça change. Platforms seemingly being happy about being regulated? Now you've got my attention.

New free EiM subscribers, it's not always like this. So, folks from Google, Spotify, Snapchat, Technology Coalition, Sal Gaming, Columbia University, Lucy Faithfull Foundation and elsewhere, don't get the wrong expectation. Do, however, become an EiM member to get full access to its archive, including our series of T&S how-to guides.

This is your Week in Review. Thanks for reading — Ben

Policies

New and emerging internet policy and online speech regulation

Almost a full 12 months since it started the process, the European Commission has formally designated OpenAI‘s ChatGPT a Very Large Online Search Engine (VLOSE) under the Digital Services Act —  a categorisation which has perplexed one or two people — while Reddit and Roblox are now Very Large Online Platforms (VLOPs).

It means each platform has passed the threshold of 45 million MAUs in Europe and will now have to comply with the toughest risk assessment and transparency obligations. With 25 other platforms now on the VLOP and VLOSE books, it's getting to be quite the workload for EU officials. 

‘A proud milestone’: Unlike Amazon (EiM #317) and Pornhub (#294) in the past, all three platforms responded to the designation with, well, something akin to delight. A Reddit spokesperson said it had been “preparing to meet VLOP requirements” — making it sound like the visit of a rarely seen relative — while Roblox said it was “proud to... reach this milestone”. Even OpenAI — who might’ve pushed back at the designation as a search engine — made no mention of an appeal and explained it was “preparing to meet the additional compliance requirements”. All very friendly, then. For now.

Also in this section...

• Would The Next George Floyd Video Survive Meta’s New Teen Safety Rules? (Techdirt)
• Our approach to online safety supervision and compliance (Ofcom)
• Lords call for AI 'kill switch' powers in UK (BBC) 
• Why the global push to break free from Big Tech keeps falling short (Rest of World)

My five takeaways from the Meta settlement

Most headlines call it a win for teenage internet users. But I don’t think that Meta setting the safety features for every major platform constitutes a success for anyone

Everything in Moderation*Alice Hunsberger

Products

Features, functionality and technology shaping online speech

Sceptics of age assurance technology have been following a story coming out of Krebs on Security, which has linked a leak of 153m US and Canadian drivers' licenses with a widely used but little known identity verification company called idscan.net, which provided services for Hertz, Target, Fedex and others. Defence secretary Pete Hegseth and ironically Brian Kerbs are in the leak, which many are claiming is a further nail in the online verification coffin.

Play the ball: I consider myself a sceptic of age assurance for the same reasons as many others; it's not particularly accurate and it's a terrible user experience. But using this debacle to dunk on online identity verification at large feels like a cheap shot. If you're asking me, the biggest problem is that LOTS and LOTS of identity data is retained in ways that it shouldn't be, whether collected online or offline. But that's a less satisfying message to shout about.

Remember when moderation systems primarily worked English and other Western languages to the detriment of users from lesser spoken languages? Well, according to Rest of World, it’s happening all over again in the context of AI.

Also in this section...

• Introducing the apgard Benchmark for Youth Sexual Safety (apgard AI)
• The Rise and Fall of Agent Civilizations (Dwarkish podcast)

Platforms

Social networks and the application of content guidelines

I’ve already mentioned OpenAI for its new designation (see: Policies) but the frontier lab also moved a bunch of year-old voluntary teen protections into a default product for 13-17 year olds. The good news is that ChatGPT for Teens includes a study mode to encourage critical thinking (read: don’t give the answer straight away) as well as stronger protections around companionship, self-harm and suicide. The bad news is that it relies on signals like account history and conversation topics to guess the age of the user, which aren't always the most robust.

Competitive advantage?: While ChatGPT for Teens apes some of Anthropic's Learning mode features that were launched last summer, positioning it as a separate product might have regulatory benefits. We know that, in the UK, "kid-safe" platforms like YouTube Kids and Lego Play are being exempted from oversight in the discussions of social media bans (EiM #343 and others). That means a standalone, clearly-labelled teen product raises the chances of a carve-out down the line.

That said, the announcement hasn't impressed everyone. Harvard professor Ryan McBain, writing for RAND, argued that the launch is “akin to a ribbon-cutting ceremony for a building that has yet to pass safety inspection”. A lot of platform safety announcements could be put into the same bracket.

Also in this section...

• US FTC Probing YouTube Over Social Media Policies (Bloomberg)
• Elon Musk Dropped the Twitter Brand. This Startup Wants to Bring It Back Without Him (Inc.com)
• Anthropic sued over alleged theft of ‘tens of thousands’ of songs (The Guardian)

People

Those impacting the future of online safety and moderation

FTC chairman Andrew Ferguson looks to be building the case he teed up in Aspen last month, when he warned that platforms "can't present one form of policy to consumers when they're deciding whether to use your platform and then have a completely different one in practice."

Bloomberg now reports the agency is in the "final stages" of a complaint against YouTube, arguing the platform's published rules say one thing while enforcement — the suspensions, the demotions, the content quietly branded violative after the fact — does another.

The framing as a consumer-protection issue about honesty rather than a speech question about content looks, on paper, like a way round the First Amendment and Section 230 — not unlike the flurry of product design cases we've seen recently. But Section 230 experts I've spoken to are sceptical that it survives contact with the actual statute. Let's see what the FTC has up its sleeve.

Posts of note

Handpicked posts that caught my eye this week

• “Really proud to launch this lab already having impacted some of the major AI models in use around the world. But we have more work to do!” - Founder and CEO Vidhya Ramalingam is bringing Moonshot's experience to AI safety.
• “When you finally get the first hard copy of your book” - Entrepreneur and friend of EiM Mark Little has a book coming out that I know many of you will love.
• “On Sept 16, will Ursula von der Leyen burnish her legacy - or will she blow it?” - Cori Crider, executive director at FOTI and recent Ctrl-Alt-Speech co-host, sets up the EU’s next big milestone.