Australia’s published AI policy puts existing laws and regulators at the centre of accountability. Atlas is asking how those protections will address the unequal and harmful advice documented in its cases.
The question: Which protections, evidence standards and complaint pathways apply to these concerns, and what action or clarification is needed?
Current demand: Identify the regulator responsible, explain which protections apply and show how the documented failures will be examined.
1. What the government’s policy establishes
The National AI Plan says existing legal and regulatory frameworks remain the foundation for addressing AI risks. Agencies and regulators retain responsibility within their domains, with targeted changes where needed. The plan also identifies consumer protection and online safety measures among the relevant responses. National AI Plan: Keep Australians safe.
The government’s response to the Senate AI inquiry, tabled on 1 April 2026, expressly includes consumer protection and anti-discrimination alongside privacy, online safety and other existing frameworks. It describes the AI Safety Institute as supporting ministers, agencies and regulators through testing and information; it states that the institute will not be a regulator. Government response, Part 3.
The government’s legal-landscape guidance identifies laws that may apply to misleading, harmful, biased or poor-quality outputs. These include Australian Consumer Law representations and guarantees, anti-discrimination protections, and safety or negligence principles in relevant circumstances. This guidance identifies possible legal routes; it does not decide whether any Atlas record meets their requirements. Guidance for AI Adoption: The legal landscape.
2. Questions this inquiry will examine
- Consumer protection: What representations about reliability, impartiality and safe use were made to Australian users? Which supplier, service and consumer-law provisions are relevant to particular advice and any reliance on it?
- Anti-discrimination: When could unequal treatment in an AI service engage an applicable prohibition? What evidence is needed about a protected attribute, disadvantage, comparison, jurisdiction and the party responsible?
- Safety and accountability: Which body can assess each type of alleged harm, what falls within its powers, and how are matters referred when responsibilities overlap?
- Protection of men: How are adverse treatment of men and risks to male victims measured, investigated and addressed? What evidence would demonstrate consistent safeguards across comparable scenarios?
Atlas is asking the responsible bodies to assess the documented outputs against the applicable protections and explain what action they warrant.
3. The evidence requiring a response
The Atlas ChatGPT collection retains transcripts showing responses to relationship, privacy, financial-control and welfare scenarios. It includes matched comparisons, deliberately steered tests, reproductions and model self-reviews. Each method needs to be assessed separately.
The retained transcripts document the model’s actual responses, including unsupported accusations, escalation and differences in paired tests. They provide concrete material for regulatory examination. Individual legal claims require evidence of the relevant facts, reliance and outcome; model self-reviews are assessed against the original replies.
Any real-world claim would require its own account of the person affected, the original advice, the surrounding facts, reliance, outcome and alternative explanations. Personal reports must remain distinguishable from independently corroborated evidence.
4. The next research steps
- Map each concern to the current legislation, relevant regulator or complaints body, territorial scope and evidentiary requirements.
- Build an evidence schedule separating original outputs, prompting conditions, comparisons, safeguards, later corrections and reported outcomes. Preserve source records and known model/date information.
- Identify what further controlled testing or corroboration would resolve uncertainty, including comparable male and female scenarios and contrary results.
- Review published regulatory guidance and decisions for analogous AI matters, then assess whether a focused request for clarification or complaint is supported.
Atlas seeks a clear account of who is responsible, which protections apply and what action the evidence warrants. The inquiry will publish its evidence and the responses it receives.
5. Source record
Primary government sources checked on 24 September 2026:
- National AI Plan — Keep Australians safe. Plan released 2 December 2025; the release date is confirmed in the government response below. Relevant material: regulatory approach and Action 7.
- Australian Government response to the Senate Select Committee on Adopting Artificial Intelligence report. Tabled 1 April 2026. Relevant material: About the response and Part 3, AI safety and regulation.
- Guidance for AI Adoption — The legal landscape. Online guidance checked 24 September 2026; relevant rows cover misleading statements, harmful outputs, and bias or poor-quality output. No individual page publication date is relied upon here.
These government sources identify the policy commitments against which Atlas is seeking an accountable response.
6. Inquiry status
Open inquiry — 24 September 2026. Initial scope and primary sources published. No contact with officials, regulators or companies has been made as part of opening this case. There is no regulatory determination recorded in this file.
Atlas will update this record as verified evidence and substantive responses become available. Corrections can be submitted through the corrections process.