I’m writing this post because Australia’s social media ban has become a case study in how evidence gets bent to fit a conclusion that was decided before the data arrived — and every time someone with a platform declares victory, that framing hardens a little further into public memory, whether it holds up or not. Documenting it as it happens is the only way I can see, from my skill set, to keep the record straight before it sets.

[sigh] So here I go. I’m really trying not to harp on here, and I do try to progress the discussion, but there are a lot of recursive loops and logics to navigate.

I opened Jon Haidt, Ravi Iyer and Zach Rausch’s latest newsletter and there it was, sitting right under the headline like a benediction: “A journey of a thousand miles begins with a single step.” Lao Tzu, they tell us — though they can’t resist a little joke first, attributing an invented second sentence (“But if that first step is hard, then you should quit”) to unnamed critics, as if anyone doubting six months of shaky data is the one being glib here.

So I went back to Chapter 64 itself, instead of the fridge-magnet version. It’s a strange passage to reach for if you’re trying to defend a law that was drafted, introduced and passed in under two weeks with a 24-hour public comment window. The chapter gives two pieces of advice that I want to pick up here, and the ban’s defenders have managed to miss both of them. The first: deal with problems while they’re small. “That which is at rest is easy to be kept hold of… break it while it is feeble, scatter it while it is small. Act before it exists, regulate before disorder.” Platform design harms didn’t arrive last December — they’ve been documented, litigated and written about for the better part of a decade, and the instrument built to address them at the design level, the Digital Duty of Care, was left to lapse while the government reached for the bluntest tool available instead. The second argument: be as careful at the end as you were at the beginning. “The common people, in their undertakings, fail on the eve of success,” the same chapter says. “If they were as prudent at the end as they are at the beginning, there would be no such failures.” That’s not an argument for patience — it’s a warning against declaring success early and getting careless right as the real test arrives. Lao Tzu, I suspect, would have had more to say about a government that let the problem grow large before acting, and commentators that want to call an unfinished job a win, than he would about critics being impatient.

That’s the kind of move I kept running into working through this. Not lies, exactly — just evidence handled the way you’d handle a fridge magnet: taken out of the room it lives in, polished up, and put somewhere it can do a different job than the one it was built for. And once I saw Haidt’s name at the top, I knew what kind of piece this was going to be…

Here’s what got me. Not that Haidt is optimistic — optimism isn’t a crime. It’s that he’s declaring victory using studies that, when you actually open them up, say close to the opposite of what he’s implying. That’s not interpretation. That’s not “reasonable people reading the same data differently.” That’s citing a source for a conclusion the source itself explicitly rejects.

I’m not going to refute this newsletter line by line. That’s not the point of this post, and honestly, it gives the piece more structural respect than it’s earned. What I want to do instead is lay out the tensions this whole saga has surfaced — the ones sitting underneath the “it’s working” narrative, that I don’t think most readers following this story casually may have had the opportunity to piece together.

The pattern isn’t new

This isn’t the first time Haidt’s relationship with evidence has been flagged. Candice Odgers, reviewing The Anxious Generation in Nature back in 2024, called him “a gifted storyteller” whose “tale is currently one searching for evidence” — noting that hundreds of researchers looking for the effects he describes have found “a mix of no, small and mixed associations.” She pointed to a 72-country analysis of nearly a million people, which found no evidence that the global rollout of social media was associated with widespread psychological harm — if anything, the associations ran mildly the other way, with higher adoption linked to slightly better wellbeing, particularly among younger users. The study’s own authors are careful about what this does and doesn’t show: the associations were small, descriptive rather than causal, and drawn from a single platform’s data at a national level — not proof that social media is good for people, just an absence of the widespread harm signal a precautionary ban would presume. Odgers still specifically flagged that age-based restrictions and device bans were “unlikely to be effective in practice — or worse, could backfire.”

Haidt was flown in anyway, as the scientific keynote. Axel Bruns — an Australian Research Council Laureate Fellow at QUT and a past president of the International Association of Internet Researchers, who attended the Sydney leg in person — called it “a curious event.” The SA premier opened proceedings by declaring “the results are in and the science is settled,” which, Bruns notes dryly, “immediately undermined the summit’s stated consultative intent.” The keynotes bore that out: Jean Twenge delivering what Bruns calls “a masterclass in casually sliding from mere appearances of vague correlation to strongly suggesting but not explicitly claiming causation,” followed by Haidt. Meanwhile, the small number of Australian scholars actually invited — from a country Bruns notes is home to some of the field’s world leaders — were relegated to breakout sessions that, unlike the keynotes, weren’t included in the livestream. His verdict: “It was a petty slap in the face of our world-leading, home-grown digital media expertise… taxpayer money was wasted on flying out professional manufacturers of concern from the United States, just because their narrative suited the predetermined policy outcomes.” Two years on, watching Haidt mark his own homework on the policy his book helped inspire, it’s hard not to see the same move happening again, just with a different dataset.

What the actual evidence says

I went back through the studies they drew on to support their position on the “misconceptions” of the critics of the ban. Here’s the shape of it:

Barnes et al., BMJ — the most methodologically rigorous thing in this entire debate: preregistered, regression discontinuity design, STROBE-compliant. Their conclusion, verbatim: “little evidence was found of immediate substantive reductions in reported social media use by adolescents under 16 years.” Their causal estimate was a statistical null (P=0.92 and P=0.60). Their own follow-up opinion piece went further, warning that “policy decisions will continue to outpace the evidence needed to inform them,” and calling explicitly for evaluations “free from industry influence.”

Bursztyn et al. — a working paper, not peer reviewed, titled, with admirable bluntness, “Why Bans Fail.” Their model finds the only stable equilibrium for compliance sits around 18%, below the 27% currently observed — meaning their own math says things are more likely to get worse, not better. Worth noting: two of the authors hold equity in a digital-wellness company whose product category is exactly what they recommend as the fix.

The eSafety Commissioner’s own parent survey — found that the single biggest reason kids retained their accounts was that the platform simply hadn’t gotten around to checking their age yet. Not sophisticated evasion. Not circumvention. Nobody asked.

Molly Rose Foundation — 61% of previously-active 12-15 year olds still had an account four months in, 70% said it was “easy,” and 60-64% said the platform had taken no action at all. Their analysis is blunt about what this means: it gives parents “a false sense of safety” while letting platforms “off the hook” for the safety-by-design work that might have actually helped.

None of this is ambiguous. None of it requires charitable reading to arrive at “not working yet, and possibly not built to work at all.”

The bit that actually made me sigh

Haidt’s newsletter lists “companies are making their products safer” as one of five reasons for optimism. I went looking for what backs that up. There isn’t anything. Not in what I could find, not in any of the studies cited alongside it. And the reason is structural, not incidental: the ban doesn’t ask platforms to change anything about how they’re built. It asks them to remove a demographic. Design, algorithms, engagement mechanics — completely untouched, for whoever remains on the platform, child or adult.

Lisa Given put this more precisely than I have: the law “does not hold technology companies to account for the content they present, or the potential harm posed by their algorithmic designs.” What has reappeared, in May 2026, is a likely to be a zombie version of the idea — a consultation paper, not legislation, reanimating some of the same design-based principles Zoe Daniel’s original bill contained, but stripped of urgency and arriving into a regulatory landscape the ban and its industry codes have already reshaped. It proposes real teeth on paper — penalties of the greater of 5% of global turnover or $50 million, risk assessments, researcher data access — but it’s still pre-drafting, with a 12-month transition period built in even after it passes, if it passes. Which puts genuine operation, at the earliest, years away.

Sequencing, not just failure

This is the part I think gets missed in the “is it working / isn’t it working” framing entirely, and it’s the point I keep circling back to. It’s not just that the ban hasn’t reduced access. It’s that while everyone argues about that, an entire age-assurance and identity-verification architecture is being built and commercially entrenched underneath it — through the ban’s enforcement apparatus, and through the Phase 2 industry codes running in parallel, developed by industry associations rather than debated in Parliament. Vendors are signing multi-year contracts. Trusted-provider lists are being drawn up. The age-verification industry itself is now lobbying eSafety for stronger powers, because a bigger, more entrenched enforcement regime is a bigger market for them.

By the time the Digital Duty of Care — the thing that would actually regulate design rather than access — arrives, if it arrives, it walks into a room where the surveillance infrastructure is already the established baseline. It doesn’t replace that infrastructure. It accommodates it.

And nobody who matters has weighed in yet

Here’s the thing that undercuts Haidt’s “it’s succeeding” framing more than any single study: the actual evaluation hasn’t happened. The government’s own assessment — led by Stanford’s Social Media Lab with an eleven-member international Academic Advisory Group — is still underway, feeding into a legislative review that doesn’t even start until 2027. Design and analysis sit with eSafety and Stanford; the advisory group’s own statement is careful to note it “does not represent the government, nor does our work constitute either endorsement or opposition to the legislation” — real independence, but advisory independence, with no lever to act on what it finds. The panel includes Amy Orben, the Cambridge researcher who co-authored the ABCD brain-imaging study that Candice Odgers cited against Haidt’s “great rewiring” thesis in her original Nature review — a preregistered study of over 10,000 children that found no meaningful relationship between screen engagement, including social media specifically, and neurodevelopment, cognition or wellbeing, “even if we set the evidential threshold very low.” Haidt is nowhere in that process. Not on the Stanford team, not on the advisory panel, not named anywhere in eSafety’s evaluation documentation. He’s commenting from outside it, ahead of it, using a working paper called “Why Bans Fail” — whose own modelling puts the stable compliance equilibrium below the rate currently observed — as evidence that it’s succeeding, while the scientists actually tasked with finding out are still years from reporting.

And in among all of it — the industry lobbying, the compliance updates, the newsletter victory laps — the people this law is actually about are almost entirely absent. When researchers did ask them: 72% of under-16s told Bursztyn’s team they’d prefer a self-limiting app to an outright ban. Molly Rose found only 31% of affected kids felt safer; 14% felt less safe. That’s not nothing. That’s the population the law claims to protect, telling anyone who’ll listen that the thing built for them wasn’t built with them.

It’s too early to call this a failure. I want to be careful about that, because “too early to tell” cuts both ways — it’s also too early to call it a success, which is exactly what Haidt did anyway. That’s the thing about being careless at the end instead of the beginning: the government rushed the law through in eight days, and now its loudest defender is rushing the verdict, declaring the journey is on the path to success if we just give it time… before the first proper measurement has even been taken. Lao Tzu had a word for that too, a few lines further into the same chapter: whoever grasps, loses.