Require deployers to keep records sufficient to reconstruct consequential AI actions
18 submissions. Each submission counts once for this shared ask. Counts do not measure importance or merit.
A shared ask groups requests for substantially the same action from three or more submissions when the group was created. Groups remain listed if withdrawals reduce their counts. AI models summarised recommendations, grouped them and checked the groups. The label summarises the requests; it is not a quote. A shared ask does not mean agreement on anything else.
Theme: Audit trails and record-keeping
Recommendations from submissions
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. Extracts are chosen by an AI model. Submissions are listed in submission number order; recommendations follow their order on each submission’s page.
Urbyte Pty Ltd Submission 48
Require execution to produce a durable receipt linked to the governing decision and source evidence.
Execution must produce a durable receipt linked to the governing decision and source evidence.
Workday Submission 51
Require trace logs sufficient to reconstruct agent actions, the principal, authority, systems used and any human approval, override or escalation.
Agents should maintain trace logging sufficient to reconstruct and audit what an agent did, on whose behalf, under what authority, using which systems, and whether human approval, override, or escalation occurred.
Attune Legal Submission 59
Require configuration retention and auditability instead of prospective mapping; deployers whose agentic systems affect people must retain operative instructions and configurations so they can answer for the system’s actions.
Require configuration retention and auditability for agentic systems, as the practical record-keeping mechanism, in place of prospective system mapping. An entity deploying an agentic system in a way that affects people should retain the instruction artefacts and configuration under which it operated, so it can answer afterwards for what the system did.
DataMPowered Pty Ltd Submission 60
For important decisions and actions, show knowledge, inference, applicable rules, authorisation, AI contributions, human judgement and pathways to review or challenge.
For important decisions and actions, institutions should be able to show what was known, what was inferred, which rules applied, who or what was authorised to act, where AI contributed, what judgement was applied and how the outcome can be reviewed or challenged.
Dr Rebecca L. Johnson Submission 107
Have the Office of AI coordinate risk-proportionate minimum evidence requirements enabling authorised investigators and evaluators to reconstruct consequential AI behaviour.
The Office of AI should coordinate the development of risk-proportionate minimum evidentiary requirements sufficient for authorised investigators and evaluators to reconstruct consequential AI behaviour.
Nuvento Submission 128
Agencies procuring AI for consequential use should specify and verify provenance retention: sources, processing, model versions and material configurations, with controlled access and retention periods sufficient for investigating outputs.
Agencies procuring AI for consequential use should specify and verify: • Evidence and provenance: retain sufficient information about sources, processing steps, model versions and material configurations to investigate an output, with controlled access and retention periods.
Infosys Submission 141
Build on the incoming Privacy Act automated decision-making transparency reforms, and require evidence-grade decision records for consequential AI-supported decisions in public service delivery.
The Australian Government should build on the incoming Privacy Act reforms regarding automated decision-making transparency obligations. The Government should require evidence-grade decision records for consequential AI-supported decisions in public service delivery.
United Nations Special Rapporteur on human rights and counter-terrorism Submission 160
Build record-keeping and audit trails into AI systems and maintain public registers of AI systems used in counter-terrorism.
The design of AI systems should incorporate adequate record-keeping capabilities, including automatic recording, in order to identify relevant events and risks and enable an audit trail. Public registers should also be maintained of AI systems used in counter-terrorism.
Cordexa Submission 230
Make the evidence record the unit of accountability: AI output informing a consequential decision should carry, in readable form, its sources, checks passed, refusals and approver.
Make the evidence record the unit of accountability. Any AI output that informs a consequential decision should carry, in a form a person or regulator can read, what sources it relied on, what checks it passed, what it refused to answer, and who approved it.
Dr Evgenii Arsentev Submission 243
Require machine-readable per-run records for Commonwealth AI use instead of policies or checklists, capturing model version, tool actions and authorisation.
For Commonwealth use of AI, require a machine-readable per-run record rather than a policy statement or a self-assessment checklist. A published policy describes an intention; a run record is evidence. The three facts that matter most — which model version acted, what it did through tools, and who or what authorised it — cost essentially nothing to capture at the time and are impossible to reconstruct afterwards.
Australian Services Union (ASU) Submission 281
Require human review and a responsible manager for every AI-made industrial-relations decision; keep records enabling independent review under the agreement where AI materially informs an employment decision affecting an employee.
There is a need for humans to review every Al- made decision in industrial relations, and for a human manager to be designated as responsible. Where AI materially informs an employment-related decision affecting an employee, the employer must keep sufficient records to enable the basis of the decision to be independently reviewed under this agreement.
A program corrected obvious text-recognition errors in this quote for display.
Australian Lawyers Alliance (ALA) Submission 310
Require claims administrators using AI to preserve inputs, outputs, system versions, decision logs and records of human intervention for the life of the claim and any applicable review, appeal or litigation period.
Entities using AI in claims administration should preserve all relevant inputs, outputs, system versions, decision logs and records of human intervention for the life of the claim and the duration of any applicable review, appeal or litigation period.
Ms Litsa Roberts Submission 352
Retain sufficient decision records to reconstruct material AI-influenced decisions, including system version, inputs, output, decision rules, accountable decision-maker, whether the output was accepted or overridden, and the outcome.
Sufficient decision records should be retained to reconstruct material decisions, including the relevant system or model version, material inputs, recommendation or output, applicable decision rules, accountable decision-maker or reviewer, whether the AI output was accepted or overridden, and the final outcome.
Mr Mark Neugebauer Submission 361
Require high-risk systems to produce tamper-resistant records sufficient to reconstruct their actions.
High-risk systems should produce tamper-resistant records sufficient to reconstruct their actions.
Name withheld Submission 477
Compliance Requirements for High-Risk Systems: 1. Pre-deployment conformity assessment – Third-party evaluation before market entry 2. Risk management system – Documented processes for hazard identification and mitigation 3. Data governance – Training data quality, provenance documentation, bias testing 4. Technical documentation – Complete model architecture, decision logic, limitations 5. Logging and traceability – Audit trail for all automated decisions 6. Human oversight – Meaningful human involvement in high-stakes decisions 7. Accuracy and robustness – Minimum performance thresholds; adversarial testing 8. Cybersecurity – Protection against prompt injection, adversarial attacks, data poisoning
Mr Martin Radzaj Submission 506
A duty on developers and deployers of frontier and agentic AI systems to maintain write-once, tamper-evident records sufficient to reconstruct an incident: model and harness specifications, system and task prompts, tool calls and outputs, network activity, sandbox and safeguard configuration, monitoring logs, and the key human decisions made.
Dr Aashish Srivastava Submission 537
For higher-consequence agentic AI, require deploying organisations to preserve reconstructable action records, identifying where relevant: objectives; permissions, credentials and tools; material tool/system interactions and external actions; human approvals; and material permission/boundary changes.
For higher-consequence agentic AI, deploying organisations should be required to preserve sufficient records to reconstruct legally consequential external actions. At a minimum, where relevant to the application, those records should identify: 1. the objective or instruction under which the agent was operating; 2. the permissions, credentials and tools available to the agent at the relevant time; 3. material tool invocations and interactions with external systems; 4. material external communications, transactions or other consequential actions; 5. applicable human approval requirements and whether those approvals occurred; and 6. material changes to permissions or operational boundaries relevant to the event.
Ms Jessica Browne Submission 555
For AI/automation affecting financial records, payroll, tax, payments, superannuation or government reporting, record relevant system/version, initiator, inputs, authority, tools/APIs, approvals, changes and actions; reject “system generated” where a precise cause is reasonably recordable.
Establish minimum evidentiary requirements for consequential AI and automation. Where AI or automation can affect financial records, payroll, taxation, payments, superannuation or government reporting, evidence should identify, where relevant, the system or model and version, initiating actor, information relied upon, authority exercised, tools or APIs used, approvals obtained, records changed and resulting actions. A description such as "system generated" should not be sufficient where a more precise technological cause can reasonably be recorded.