New laws are being proposed proposed by the Albanese government to crack down on social media platforms enforcement of the social media age restrictions. The current provisions are seriously flawed, in that they are intrusive (even a child could be forced to provide information) and they do not advance transparency.
We propose four amendments to the Bill to fix the issues in question.
We are calling for genuine, complete transparency from large social media platforms — not the selective schemes used by the EU and Ofcom, where content itself is never published and censorship becomes hidden. Our position is that the only appropriate approach is full transparency, including provision of source code and complete algorithms.
This matters as much for child protection as for free speech. The "reasonable steps" test is a proportionality test: the riskier the platform, the stronger the measures needed to keep children off it, and that risk cannot be assessed without visibility into how a platform operates. Our analysis of the eSafety Commissioner's own trial data shows why: recommended age-assurance systems let over 40% of 10-year-olds through the age gate, and some misjudged Indigenous children as being in their 50s and 60s. We should all be able to know if we are being psychologically manipulated.
The same logic applies to feed and moderation algorithms and to psychological experimentation. Platforms have a documented history of running unethical experiments (such as Meta's infamous emotional contagion study) on unwitting users, including children, to shape design decisions affecting their wellbeing. This should also be published.
Division 3—Transparency obligations
Algorithms affecting distribution of material
(1) The following items must be published by a large social media platform or any subcontractor.
(a) the source code of each algorithmic system used materially to determine for any Australian end user:
(i) the selection of material or content presented to an end-user;
(ii) the ranking or ordering of material or content;
(iii) the recommendation of material, content or accounts;
(iv) the prevalence or prominence of material, content or accounts;
(v) whether that end-user is more or less likely to be shown particular material, content or accounts; or
(vi) the selection or ranking of search results;
(b) for each artificial intelligence or machine-learning model used for a purpose mentioned in paragraph (a):
(i) model weights;
(ii) system parameters; and
(iii) documentation sufficient to explain the operation, training and deployment of the model as to enable empirical experimentation on the system (including by making modifications);
Content moderation systems
(c) the source code, model weights and system parameters of each automated system used materially for any Australian end user to:
(i) identify or classify material for moderation;
(ii) remove material;
(iii) restrict or disable access to material, content, or accounts;
(iv) suspend, restrict or terminate an account;
(v) alter the prevalence or prominence of material; or
(vi) refer material or an account for human moderation;
(d) the policies, rules, criteria, instructions and decision-making processes governing:
(i) content moderation;
(ii) account moderation;
(iii) automated moderation;
(iv) human moderation; and
(v) review or appeal of moderation decisions;
Age-assurance systems
(e) for each system used to determine, estimate, infer, verify or assure the age of any Australian end-user:
(i) its source code;
(ii) any model weights;
(iii) its system parameters;
(iv) the categories of information and data relied upon;
(v) the processes (including decision rules and/or thresholds) by which an end-user is referred to another method of age assurance; and
(vi) documentation sufficient to explain the operation of the system and the circumstances in which an end-user may be required or requested to provide identification or additional information and to enable empirical experimentation on the system (including by making modifications).
Moderation and Censorship Data
(f) Word clouds in the form of phrase and frequency count, banded by hour of each of the following:
(i) posts complained about to the platform.
(ii) posts subject to any substantive moderation action by the platform (e.g. removal, or reducing reach).
(iii) posts generally on the platform
Psychological Experiments on End Users
(g) The results and procedure of any and all human subject experiments (including A/B testing) conducted on end users of any social media platform or other interactive system (regardless of geographical location or other trait of the end users studied) which has been considered when designing the algorithms, policies, platforms and processes identified in this section.
(h) For subsection (g), it is irrelevant whether an experiment was conducted or conceived (either in full or in part) prior to the commencement of this provision, or that this was conducted by another person.
Note: More detailed amendment text is available on request from the Free Speech Union of Australia.
We are calling for Section 220 to be amended to provide a right of review over any information request made by the eSafety Commissioner, available to any natural person served with a notice.
The power as drafted reaches any person, yet carries no review right. The absence of merits review matters more, not less, where the power is broad in character: a narrow power is self-constrained, a broad one depends on external checks. That self-restraint cannot be assumed — in X Corp and eSafety Commissioner [2025] ARTA 2230, the Tribunal set aside an exercise of the Commissioner's existing powers, and that misuse only surfaced because a review right existed. The same is true of Baumgarten.
Section 220 Online Safety Act 2021
After subsection 220(1), insert:
Information-gathering decisions
(1A) An application may be made to the Administrative Review Tribunal for review of a decision of the Commissioner to give a natural person a notice under a provision of this Act requiring the person:
(a) to give the Commissioner information; or
(b) to produce to the Commissioner a document; or
(c) to give the Commissioner information and produce to the Commissioner a document.
(1B) An application under subsection (1A) may be made by the natural person to whom the notice was given.
(1C) For the avoidance of doubt, on review of a decision referred to in subsection (1A), the Administrative Review Tribunal may exercise all of the powers conferred on the Tribunal by the Administrative Review Tribunal Act 2024, including by considering whether:
(a) the giving of the notice was necessary and proportionate;
(b) the information or documents required by the notice are relevant to the performance of a function or exercise of a power under this Act; and
(c) the scope, burden and period for compliance with the notice are reasonable.
After Section 220 Online Safety Act 2021
Insert:
220A Effect of application for review of information-gathering decision
(1) If an application is made under subsection 220(1A), the operation of the decision is suspended until the application is finally determined or otherwise disposed of.
(2) Subsection (1) does not prevent the Commissioner from applying to the Administrative Review Tribunal for an order that the decision, or a specified part of the decision, continue to operate.
(3) The Tribunal may make an order under subsection (2) if satisfied that it is necessary in the circumstances.
We are calling for enforcement of the Basic Online Safety Expectations and the Social Media Minimum Age (and any future Digital Duty of Care) to be moved to the Australian Competition and Consumer Commission (ACCC).
The many hats the eSafety Commissioner wears are unseemly — there is no separation between rule-maker and enforcer. This creates a risk of selective enforcement which has unfortunately been realised.
Unlike eSafety, the ACCS has the institutional experience, litigation capability and technical resources the Commissioner's office has shown it lacks. By contrast, eSafety has uncritically adopted a deeply flawed technical trial containing AI-hallucinated citations and recommended systems that let over 40% of ten-year-olds through an age gate. As was reported in Parliament, she has also used Professor Jeff Hancock to chair her 'research advisory board' evaluating the ban, even though he previous undertook psychological manipulation experiments on Facebook users without ethical approval.
Not only does this protect from censorship, this also means that the social media ban can be enforced.
The enforcement powers in sections 50, 53, 57, 60, 63D, 63DA, 63DB, 63H of the Online Safety Act 2021 are not enforceable by the Commissioner, but instead are enforceable by the ACCC under Part 4 of the Investigatory Powers Act.
Note: More detailed amendment text is available on request from the Free Speech Union of Australia.
We are calling for a statutory prohibition on the eSafety Commissioner, her staff, and those acting on her behalf using informal pressure (or "jawboning") to achieve outcomes she has no lawful power to achieve.
The Commissioner's response to Baumgarten, maintaining she retains the power to issue informal notices and treating the case as a one-off rather than accepting any general principle, shows why a clear prohibition, with criminal consequences for intentional or reckless breach, is now necessary.
28A Commissioner must not engage in informal enforcement action
(1) The Commissioner must not request, encourage, pressure, induce or otherwise seek to procure a provider of a social media service, relevant electronic service, designated internet service or internet search engine service with the purpose or effect of:
(a) removing material; or
(b) restricting, suppressing, demoting, de-indexing or disabling access to material; or
(c) suspending, restricting or terminating an end-user's account; or
(d) otherwise restricting the publication, availability, visibility or dissemination of material;
otherwise than in the exercise of a power expressly conferred on the Commissioner by this Act or another Act.
(2) A person commits an offence if:
(a) the person is the Commissioner, a member of the staff of the Commissioner, or a person acting on behalf of the Commissioner (whether or not authorised to do so); and
(b) the person intentionally or recklessly engages in conduct prohibited by subsection (1).
Penalty: Imprisonment for 2 years or 120 penalty units, or both.
(3) For the purposes of subsection (1), it is immaterial whether:
(a) the communication is described as an informal notice, informal request, complaint alert, referral, notification, request for voluntary action or otherwise; or
(b) the communication states that compliance is voluntary; or
(c) the provider is legally required to comply with the communication; or
(d) the provider takes the action sought.
(4) Subsection (1) does not prevent the Commissioner from notifying a provider of the existence of a complaint, provided that the Commissioner does not request, encourage or seek to procure action of a kind mentioned in paragraphs (1)(a) to (d).
Join us in fighting for free speech and protecting this vital liberty. You can become a member for as little as $5 a month. Membership is usually tax-deductable, too.