In May 2026, Australia's eSafety Commissioner sits at the centre of the most ambitious — and most contested — online safety experiment in the democratic world. Five months into enforcing the under-16 social media minimum age law that took effect in December 2025, and with Phase 2 industry codes now reaching into generative AI and search engines under the Basic Online Safety Expectations (BOSE), Canberra is running an experiment whose results will shape regulation from Sacramento to Brussels.
We support clear, proportionate online safety rules. We are increasingly worried that Australia's framework — born of genuine concern about youth wellbeing — is drifting toward a model that treats the internet less like an open commons and more like a licensed broadcaster.
What's actually being enforced
The Online Safety Amendment (Social Media Minimum Age) Act 2024 requires age-restricted platforms — Meta's services, TikTok, Snapchat, X and similar — to take "reasonable steps" to prevent Australians under 16 from holding accounts. Penalties reach AUD 49.5 million per breach. Since the December 2025 commencement, eSafety has been publishing compliance guidance and signalling that age-assurance technology — not just self-declaration — is expected.
Simultaneously, Phase 2 of the industry codes process under the Online Safety Act 2021 has moved beyond the original eight sectors. Draft codes now cover generative AI services and search engines, requiring measures against synthetic child sexual abuse material, non-consensual intimate imagery, and "class 1" content. If industry codes are deemed inadequate, eSafety can impose binding standards — a power it has already used.
The case for the regulator
None of this came from nowhere. Australian parents and clinicians have raised legitimate alarms about adolescent mental health, algorithmic amplification of self-harm content, and the speed at which generative tools can produce realistic abuse material. Commissioner Julie Inman Grant has consistently argued that doing nothing while these harms scale is not a neutral choice.
That framing is gaining international traction. As MediaNama reported on 14 May 2026, OpenAI has now publicly endorsed the bipartisan US Kids Online Safety Act and a frontier-AI safety bill in Illinois — arguing AI "must not repeat social media's mistakes." When the largest AI developer asks for legislation, the political cost of regulatory restraint shrinks.
Where proportionality breaks down
The problem is not the goal. It is the mechanism. Three concerns deserve more weight than they currently get in Canberra's debate:
- Age assurance imposes adult costs. A minimum-age rule that platforms can actually verify means, in practice, age-checking everyone. The Australian government's own age assurance trial reported in 2025 that no current technology achieves both high accuracy and strong privacy without trade-offs. Document uploads, face scans, or inference-based estimation each create new surveillance surfaces — risks that fall on the entire adult population, not just minors.
- Teenagers don't disappear; they migrate. The under-16 ban applies to designated platforms. It does not apply to messaging apps, gaming chat, or open-web forums. Pushing teenagers off moderated mainstream services and toward less-resourced alternatives is not obviously a safety win.
- The codes process is opaque by design. Industry codes are negotiated, but binding standards can be imposed unilaterally. For generative AI — a sector where the regulatory perimeter is still being drawn worldwide — Australian standards risk becoming a de facto global compliance floor for any company that wants to serve Australian users.
The civil-liberties critique is not fringe. The Electronic Frontier Foundation has spent 2026 cataloguing how social-media age laws in the US replicate the same trade-offs Australia is now operationalising at scale: surveillance creep, speech chilling, and the displacement of harms rather than their reduction. Australian digital rights advocates including Lizzie O'Shea have raised analogous concerns about the speech costs of broad platform duties.
A better path: outcomes, not gatekeeping
There is a more proportionate version of this agenda — and Australia is well-placed to lead it.
First, regulate design rather than access. Default privacy settings for minors, restrictions on engagement-maximising recommender systems for adolescent accounts, and transparency obligations around algorithmic ranking achieve much of the public-interest goal without forcing universal age verification. The UK's Age Appropriate Design Code is a useful template.
Second, narrow the scope of binding AI codes to genuinely catastrophic harms — non-consensual intimate imagery, CSAM, terrorism content — and use co-regulation rather than top-down standards for everything else. Frontier AI safety bills like those endorsed by OpenAI focus on incident reporting and transparency, not content prohibitions; that distinction matters.
Third, build in real sunset and review clauses. The under-16 ban includes a statutory review, but the global signal Australia is sending — that minimum-age laws are workable — will outrun any domestic reassessment. Publishing rigorous, independent evidence on actual youth outcomes by mid-2027 should be non-negotiable.
Why this matters beyond Australia
Other jurisdictions are watching. The EU is calibrating Digital Services Act enforcement against minors; the UK is implementing Ofcom's Online Safety Act codes; California's social-media restrictions face constitutional challenge; India is drafting safe-harbour reforms. If Australia's model is judged successful, the template — a powerful unilateral regulator, broad ministerial powers, and access-based rules — will travel.
The right lesson from Canberra's experiment is not that online safety regulation should be abandoned. It is that the regulator's tools should match the problem's actual shape: targeted, evidence-led, design-focused, and humble about what enforcement can achieve. Australia has world-class institutional capacity. The next twelve months will show whether it uses that capacity to set a global standard for proportionate safety regulation — or for the opposite.