AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 339

Mr Peter Henss

Submitter group: Individuals.

Position labels are not shown for individual submitters.

An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.

Open in the interactive guide

Summary

Read the original submission for the full text.

Proposes a mandatory disclosure and audit obligation for recommendation feeds and for AI systems whose internal safety judgements determine what users or agencies are told. Describes a Commonwealth Bill the submitter drafted and published in a companion volume (attention levy, civic content quota, auditor-general function, portability standards, Indigenous data provisions); notes one provision is not yet rights-compatible.

Key points

Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.

  1. States that no general obligation currently requires feed operators or AI providers to disclose the criteria or exclusions applied when shaping what users see.

    No general obligation currently requires a recommendation-feed operator or an AI provider to disclose the criteria, or the exclusions, applied when shaping what a user sees.

  2. Says infinite scroll, autoplay and engagement-optimised ranking are treated as protected design choices, and that the Bill would reclassify them as design defects under Australian Consumer Law.

    Infinite scroll, autoplay, and engagement-optimised ranking are currently treated as protected design choices. The Bill reclassifies them under the Australian Consumer Law as mechanical cognitive design defects

  3. Argues that existing Commonwealth AI controls govern process but do not require disclosure of what an AI model itself excludes.

    The Policy for the Responsible Use of AI in Government, updated in December 2025, requires agencies to complete AI impact assessments, publish AI Transparency Statements, and name Accountable Officials. These controls govern process. None require disclosure of what an AI model itself declines to include during synthesis.

  4. Argues there is a documented relationship between engagement-optimised feeds and harm to younger and vulnerable users.

    This responds to the documented relationship between engagement-optimised feeds and harm to younger and vulnerable users, including content correlated with disordered eating, gambling-adjacent design, and compulsive use.

  5. Describes a Bill provision letting a First Nations governance council order unlearning where models ingested cultural material without consent.

    Where a model has ingested First Nations cultural material without consent, the Bill empowers a First Nations Data and AI Governance Council to issue binding statutory unlearning orders.

  6. Describes an attention levy that would treat engagement-optimised amplification as a negative externality and price it rather than ban it.

    the levy treats engagement-optimised algorithmic amplification as a negative externality and prices it accordingly, rather than banning it outright.

Recommendations

Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.

  1. Introduce mandatory disclosure and independent audit for recommendation-feed and AI safety decisions shaping what users or agencies see; log those decisions without regulating content or determining truth.

    I propose that Australia introduce a mandatory disclosure and audit obligation for two categories of algorithmic decision-making: recommendation feeds that determine what content spreads, and AI systems whose internal safety judgements determine what a user or an agency is told. Where either system makes a decision that shapes what someone sees, that decision should be logged and available to independent audit. This is not a proposal to regulate content. It does not ask any platform, developer, or regulator to decide what is true. It requires disclosure of the mechanism, not control of the message.

  2. Consider five legislative tiers, from minimal disclosure to an audit-authority model, matching intervention to the evidence.

    I have drafted this proposal as a working Commonwealth Bill, the Digital Cognitive Infrastructure Act 2026, together with two companion Acts required under section 55 of the Constitution, a Statement of Compatibility with Human Rights, an Explanatory Memorandum, and a Second Reading Speech. It is drafted at five tiers, from a minimal disclosure-only requirement to a structurally ambitious audit-authority model, so the Committee can consider the level of intervention that matches the evidence before it.

  3. Price engagement-optimised algorithmic amplification through an attention levy, using separate Imposition and Assessment Acts, and direct revenue through the Consolidated Revenue Fund to a Cognitive Harm Compensation Scheme.

    Modelled on the economic logic that phased out leaded petrol, the levy treats engagement- optimised algorithmic amplification as a negative externality and prices it accordingly, rather than banning it outright. It is structured as two separate bills, an Imposition Act and an Assessment Act, to satisfy the single-subject rule in section 55 of the Constitution. Revenue is directed through the Consolidated Revenue Fund to a dedicated Cognitive Harm Compensation Scheme.

  4. As part of the proposed Bill, require platforms to allocate a fixed proportion of feed impressions to verified civic, educational and public-interest content.

    The Bill extends the same logic to recommendation feeds, requiring platforms to allocate a fixed proportion of feed impressions to verified civic, educational, and public-interest content.

  5. Establish an independent Algorithmic Auditor-General function through the existing Digital Platform Regulators Forum, instead of a new centralised regulator.

    Rather than create a new centralised regulator, the Bill deliberately rejects that model. An earlier draft mechanism proposing a standalone Cognitive Infrastructure Authority was considered and set aside during drafting, on the assessment that a new mega-regulator would duplicate existing powers and risk a chilling effect on legitimate innovation. In its place, the Bill establishes an audit function modelled on the statutory independence of the Auditor-General, working through the existing Digital Platform Regulators Forum rather than beside it.

  6. Amend the Consumer Data Right to require platform preference data in a standardised public reference format, allowing competing services or independent middleware to interoperate without accessing trade-secret model weights.

    The Bill amends the Consumer Data Right regime in the Competition and Consumer Act 2010 to require platforms to publish preference data against a standardised public reference format, so a competing service, or independent middleware, can genuinely interoperate without accessing a platform's trade-secret model weights.

  7. Require Indigenous Cultural Asset Scans before AI Model Register listing and compliance with Traditional Knowledge labels; empower a First Nations Data and AI Governance Council to order unlearning of cultural material ingested without consent.

    The Bill requires AI developers to complete an Indigenous Cultural Asset Scan before listing a model on the mandatory Australian AI Model Register, and to obey embedded Traditional Knowledge labels drawn from the Local Contexts framework. Where a model has ingested First Nations cultural material without consent, the Bill empowers a First Nations Data and AI Governance Council to issue binding statutory unlearning orders.

  8. Reclassify infinite scroll, autoplay and engagement-optimised ranking as mechanical cognitive design defects under Australian Consumer Law.

    Infinite scroll, autoplay, and engagement-optimised ranking are currently treated as protected design choices. The Bill reclassifies them under the Australian Consumer Law as mechanical cognitive design defects, using the same logic that once reframed unsafe vehicle design as a product liability question rather than a matter of driver error.

  9. Require AI used in government research or advisory pipelines to record verified material it excluded and the basis for exclusion.

    The Bill's audit provisions, Part Nine, require an AI system used in a government research or advisory pipeline to produce a record of material it verified but excluded, and the basis for exclusion.

  10. As part of the proposed Bill, require disclosure of the optimisation objective applied to a recommendation feed, with a user option to switch to a chronological or user-directed feed.

    The Bill's feed provisions, Part Six, require disclosure of the optimisation objective applied to a recommendation feed and a user-facing option to switch to a chronological or user-directed alternative.

  11. Consider whether the AI Safety Institute's mandate should extend to verifying industry-submitted disclosure logs directly, rather than relying on self-reported compliance.

    The Committee may wish to consider whether AISI's mandate should extend to verifying industry-submitted disclosure logs directly, rather than relying on self-reported compliance.

  12. Scrutinise the proposed Part Fifteen licence-suspension power before legislation proceeds, given the accompanying human-rights statement’s finding that it is not yet compatible.

    Its Statement of Compatibility with Human Rights reaches a considered finding of not yet compatible on one provision, the licence suspension power in Part Fifteen, Enforcement. I flag this directly, as an example of the kind of enforcement power that warrants further scrutiny before any legislative process proceeds.

  13. Have the Committee recommend, in its final report, government examination of disclosure and audit obligations for recommendation feeds and AI guardrail decisions, based on the accompanying Bill’s tiered model.

    That the Committee recommend, in its final report, that the Government examine a legislative disclosure and audit obligation for recommendation-feed algorithms and AI guardrail decision-making, built on the tiered model set out in the accompanying Bill, as a direct response to the gap identified under paragraph (i) of these Terms of Reference.

Read the original

Open in the interactive guide