Submission 501
Mr Benjamin Macindoe
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.
Summary
Read the original submission for the full text.
Argues that, to the best of the submitter’s research, Australian law lacks tamper-evident recording duties for agentic AI. Proposes seven procurement questions, an AI Safety Institute pilot and a risk-scoped 2027 duty only if definable, affordable and enforceable. Discloses a separate project advocating a legal category for AI, which this submission does not advance, and asks for no authorisation regime.
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.
Argues that, to the best of the submitter's research, no Australian law requires suppliers or operators of agentic AI to keep tamper-evident records of their actions or state changes, or to make them available to an independent party.
To the best of my research, no Australian law requires the supplier or operator of such a system to keep a tamper-evident record of what it did, or of material changes to its persistent state, or to make that record available to an independent party.
States that the Digital Transformation Agency and the Australian Signals Directorate already direct Commonwealth agencies to keep such records, and that nothing yet binds the suppliers who hold them.
Two Commonwealth technical authorities, the Digital Transformation Agency and the Australian Signals Directorate, already direct Commonwealth agencies to keep such records and advise buyers to ask for them. Nothing yet binds the suppliers who hold the records to keep them honestly, or to hand them over.
Recommends that agentic AI procurement guidance and impact assessments require suppliers to answer seven written questions about records before contract and at renewal.
for agentic AI use cases, the Digital Transformation Agency's Guidance on AI procurement in government and its checklist, and the AI impact assessment required under the Policy for the responsible use of AI in government, require suppliers to answer in writing, before contract and again at renewal, seven questions about records
Proposes a staged approach: procurement conditions now, a piloted technical standard next, and a statutory duty in the 2027 standards only if a pilot shows one is definable, affordable and enforceable.
procurement conditions now, where the Commonwealth is already the buyer; a piloted technical standard next; a statutory duty in the 2027 standards only if the pilot shows one is definable, affordable and enforceable.
Argues that a regulator that can only ask an operator for its own account is working from a submission, not a record.
A regulator that can only ask the operator for the operator's own account is working from a submission, not a record, whatever its compulsory powers.
Says any recording requirement should be scoped by consequential use, autonomy and access to critical systems or data rather than by firm size, with responsibility resting on identified persons.
Any such requirement should be scoped by consequential use, autonomy and access to critical systems or data rather than by firm size, and responsibility for compliance should rest with identified natural or legal persons at all points.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
For agentic AI use cases, have the DTA's AI procurement guidance and checklist and the AI impact assessment require suppliers to answer seven written questions about records before contract and at renewal.
for agentic AI use cases, the Digital Transformation Agency's Guidance on AI procurement in government and its checklist, and the AI impact assessment required under the Policy for the responsible use of AI in government, require suppliers to answer in writing, before contract and again at renewal, seven questions about records
Ask about supplier updates, staff instructions and learned changes; recording who, when and what; supplier alteration of records; agency-specific versus global changes; traces of pauses and corrections; verifying switch-off across connections; and independent record access after problems.
1. How does the system change in normal use: updates from the supplier, instructions from staff, and things it picks up from the people and systems it deals with? 2. Is every such change recorded, with who, when and what? 3. Can the supplier quietly alter that record? 4. Did a given change affect this agency's deployment only, or every customer's? 5. Do pauses and corrections leave a trace in the same record as the fault? 6. When the system is switched off, how does the agency verify it is off everywhere it was connected? 7. If something goes wrong, can an independent party obtain the record without going through the supplier?
Record supplier answers against the use case in the agency's internal AI register, and record any blank as a finding.
The answers should be recorded against the use case in the agency's internal AI register. A supplier who cannot answer has told the agency something, and the blank should be recorded as a finding.
For agentic use cases, have contracts secure an agency-held copy of the record on a schedule, in a form showing later alteration, and the agency's right to release it to a regulator, auditor or insurer on its own authority.
For agentic use cases, contracts should secure an agency-held copy of the record on a schedule, in a form that shows any later alteration, and the agency's right to release it to a regulator, auditor or insurer on its own authority.
Identify as a gap in existing law the absence of any requirement that suppliers or operators of AI systems acting on external systems and data keep a tamper-evident record of those actions and of material changes to persistent state.
That the committee identify, as a gap in existing law, the absence of any requirement that the supplier or operator of an AI system which acts on external systems and data keep a tamper-evident record of those actions and of material changes to the system's persistent state
Consider a risk-scoped recording requirement with retention and independent-access terms for the Australian AI standards to be legislated in early 2027, if the pilot shows it definable, affordable and enforceable.
recommend that a risk-scoped recording requirement with retention and independent-access terms be considered for inclusion in the Australian AI standards to be legislated in early 2027, if the pilot in recommendation 3 shows it to be definable, affordable and enforceable.
Scope any such requirement by consequential use, autonomy and access to critical systems or data rather than firm size, with compliance resting on identified natural or legal persons.
Any such requirement should be scoped by consequential use, autonomy and access to critical systems or data rather than by firm size, and responsibility for compliance should rest with identified natural or legal persons at all points.
Make the operator's duty include ensuring that a system's modification of its own state is recorded.
Where a system modifies its own state, the operator's duty should include ensuring that the modification is recorded, because it is the one change no other party can attest to.
Commission the AI Safety Institute to develop and pilot the recording standard in consultation with CISC, OAIC, ASD, DTA, operators and civil society.
That the Government commission the Australian AI Safety Institute to develop and pilot the recording standard, in consultation with the Cyber and Infrastructure Security Centre, the Office of the Australian Information Commissioner, the Australian Signals Directorate, the Digital Transformation Agency, operators and civil society
Have the AI Safety Institute maintain a short published logging standard that agencies and small businesses can point to when putting the seven questions to suppliers.
that the Institute maintain a short published logging standard that agencies and small businesses can point to when they put the questions in recommendation 1 to a supplier.
Have the AI Safety Institute's pilot test scope and completeness, interaction with privacy law (including APP 11.2 destruction obligations), retention and access rules, cost, enforceability and statutory home.
The pilot should test scope and completeness, interaction with privacy law (including the destruction obligations in Australian Privacy Principle 11.2), retention and access rules, cost, enforceability and statutory home.
Draft any definitional or liability provisions arising from the committee's recommendations as current allocations subject to periodic review, not permanent characterisations of what AI systems are.
That definitional or liability provisions arising from the committee's recommendations be drafted as current allocations subject to periodic review, rather than as permanent characterisations of what AI systems are.