Submission 467
Mr Edward Burrowes
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.
A personal submission from the head of operations at an AI-evaluation firm argues Australia does not need a general AI Act but should build compute, attach national interest conditions to AI training permits and data centre approvals, give the AI Safety Institute a narrow information-gathering power, test Commonwealth AI agents, and close three AI-enabled biosecurity gaps.
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 Australia is about to legislate binding standards on AI inputs while the capability layer that generates catastrophic risk remains voluntary.
Australia is about to legislate binding standards on the inputs to AI, being data centre energy, water, land and copyright, while the capability layer that generates catastrophic risk remains entirely voluntary. That asymmetry is backwards, and it is also an opportunity.
Argues that Australia's advantage is trust (a treaty ally with renewable resources and abundant land), not being cheap.
Australia's equivalent advantage is trust. It is a treaty ally with no China-hedging complications, it has world-class renewable resources and abundant land, and it is already a significant destination for data centre investment. Trusted, green and sovereign is a real position. Cheap is not.
Argues that training a frontier model on Australian copyright material is prima facie infringing, which it calls a negotiating position and the second scarce input Australia controls.
Training a frontier model on Australian copyright material is therefore prima facie infringing. That is not a bug to be fixed. It is a negotiating position, and it is the second scarce input Australia controls.
Argues that the AI Safety Institute has no compulsory information gathering power or known pre-deployment testing agreement, and that its cooperative memoranda grant no access rights.
It has no compulsory information gathering power and, so far as is publicly discoverable, no pre-deployment testing agreement with any frontier developer. The Anthropic memorandum of understanding and the Australia-UK memorandum of 25 May 2026 are cooperative instruments that grant no access rights.
Argues that an agency procuring on capability benchmarks alone is procuring an unmeasured security posture.
An agency that procures on capability benchmarks alone is procuring an unmeasured security posture.
States that Australia does not require imported synthetic nucleic acids to come from providers that screen orders, and that the Director of Biosecurity could impose such a condition on import permits without new legislation.
Australia does not require imported synthetic nucleic acids to be sourced from a provider that screens orders for sequences of concern. The Director of Biosecurity can impose that condition on BICON import permits under the Biosecurity Act 2015 immediately and without new legislation.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Set an explicit national target for sovereign and allied compute capacity well above the current pipeline, and resource the transmission, firming and connection reform needed; treat firming reliability and cool-climate siting as binding constraints and do not trade away the renewable requirement.
Set an explicit national target for sovereign and allied compute capacity well above the current pipeline, and resource the transmission, firming and connection reform needed to deliver it. Treat firming reliability and cool-climate siting as the binding engineering constraints. The renewable requirement is affordable and should not be traded away.
In early-2027 AI standards, condition training permits/large data-centre approvals on AISI pre-deployment access to models trained or served at scale in Australia, serious-incident reporting and minimum weight security, priced into permits/approvals.
The AI standards legislation planned for early 2027 should make any AI training permit, and any large scale data centre approval, conditional on three national interest requirements: pre-deployment evaluation access for the Australian AI Safety Institute to models trained or served at scale in Australia; mandatory reporting of serious AI incidents; and a published minimum standard for model weight security. These conditions should be priced into permits and approvals rather than established as a separate regulatory regime.
A program corrected obvious text-recognition errors in this quote for display.
Direct DFAT/DISR standing negotiations with democratic middle powers holding AI supply-chain chokepoints to harmonise access conditions, first testing through official scenario exercises before seeking any commitment.
Direct DFAT and DISR to open a standing negotiation with the other democratic middle powers holding AI supply chain chokepoints, with the objective of harmonising the access conditions each attaches to its own chokepoint. Coordinated conditions cannot be arbitraged; unilateral ones can. This should be tested first through scenario exercises with participating officials before any commitment is sought.
Give the AI Safety Institute a narrow statutory power to require evaluation results and incident information from developers and deployers of frontier AI offered in Australia, with a mandatory serious incident reporting scheme.
Give the AI Safety Institute a narrow statutory information gathering power, being a notice to produce evaluation results and incident information from developers and deployers of frontier AI systems offered in Australia, together with a mandatory serious incident reporting scheme.
Require Commonwealth agencies deploying AI agents with tool or data access to test for prompt injection and agent hijacking before deployment and publish results in a standard form, making adversarial robustness a procurement gate.
Require Commonwealth departments and agencies deploying AI agents with tool or data access to test for prompt injection and agent hijacking specifically, not general capability, before deployment, and to publish results in a standard form. Adversarial robustness should be a procurement gate.
Have the Agriculture, Fisheries and Forestry Minister direct Biosecurity’s Director to impose sequence-screening and customer-verification conditions on synthetic-nucleic-acid BICON imports this year, without legislation.
Recommend that the Minister for Agriculture, Fisheries and Forestry direct the Director of Biosecurity to impose sequence screening and customer verification conditions on BICON import permits for synthetic nucleic acids. This requires no legislation and can be done this year.
Require the Australian CDC to publish the National Wastewater Surveillance Program's sentinel site count, pathogen panel and methodology, and fund a pathogen-agnostic metagenomic capability alongside it.
Recommendation 7. Require the Australian CDC to publish the National Wastewater Surveillance Program's sentinel site count, pathogen panel and analytical methodology, and fund a pathogen-agnostic metagenomic capability alongside the existing targeted program.
Add reusable elastomeric respirators and filters for essential and critical infrastructure workers to the National Medical Stockpile with a published rotation and readiness standard; implement Action 6 of the COVID-19 Response Inquiry including outstanding ANAO recommendations, and report readiness against a measurable target.
Recommendation 8. Add reusable elastomeric respirators and filters, sized for essential and critical infrastructure workers rather than health workers alone, to the National Medical Stockpile, with a published rotation and readiness standard. Implement Action 6 of the COVID-19 Response Inquiry, including the outstanding ANAO recommendations, and report stockpile readiness against a measurable target.
Seek on notice which Commonwealth agency owns AI-enabled biological risk, and on what budget.
Recommendation 9. Seek on notice a single answer to the question of which Commonwealth agency owns AI- enabled biological risk, and on what budget.
Read the original
In some quotes a program corrected obvious text-recognition errors for display.