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 258

GovTech Australia

Submitter group: Companies.

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.

GovTech Australia, a company working with Australian technology businesses, argues that procurement excludes them from production despite successful pilots. It proposes a sovereign GovAI lane, paid pilot pathways, a capability test and tested Australian continuity options. It supports global technology and competition, with Australian control proportionate to workload sensitivity, without guaranteed contracts or unconditional Australian purchases.

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. Argues that Australian companies can validate government pilots yet find production procurement routed through channels where they cannot practically compete.

    These initiatives are not yet connected to an equivalent production pathway. Companies can demonstrate technology and validate a government use case, only to find the production requirement channelled through an incumbent, hyperscaler, whole-of-government arrangement or procurement mechanism through which they cannot practically compete.

  2. Argues that, as government depends on AI, material Commonwealth capability should not develop a single external point of denial.

    As government becomes operationally dependent on AI, Australia should not allow material Commonwealth capability to develop a single external point of denial.

  3. Argues that sovereignty is a matter of degree, with Australian control increased by layer according to workload sensitivity.

    Sovereignty is not binary. Government should be able to increase Australian control at each layer according to workload sensitivity and consequence.

  4. Argues that government pilots should not use Australian SMEs to de-risk requirements later procured from pathways that exclude them.

    Government innovation programs should not inadvertently use Australian SMEs to de-risk requirements that are subsequently procured through pathways from which those SMEs are effectively excluded.

  5. States that no equivalent Australian sovereign alternative is presently visible in the published GovAI model brokerage offering.

    No equivalent Australian sovereign alternative is presently visible in the published brokerage offering.

  6. Argues that data residency is not a proxy for sovereignty, since foreign-controlled services hosted in Australia can remain subject to offshore control.

    Data residency is important but is not a proxy for sovereignty. A foreign-controlled service hosted in Australia may satisfy residency requirements while remaining subject to offshore corporate control, legal obligations or operational dependencies.

Recommendations

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

  1. Establish a sovereign GovAI lane alongside international frontier capability, enabling agencies to discover, test, assure and procure Australian capability across relevant stack layers.

    The published Model Brokerage provides access to international model families through major foreign- controlled provider ecosystems.[2] Its model-agnostic design is useful because applications can switch between approved models without rebuilding integrations. But model agnosticism is not the same as sovereign choice. No equivalent Australian sovereign alternative is presently visible in the published brokerage offering. GovAI should establish a sovereign lane through which agencies can discover, test, assure and procure Australian capability across relevant layers of the stack.

  2. Create a Pilot-to-Production pathway, normally paying for substantive government-requested SME proofs of concept.

    Create a Pilot-to-Production pathway, with substantive government-requested SME proofs of concept normally paid.

  3. Before substantive pilots, identify success criteria, funding, production assurance requirements, likely procurement routes and whether participating companies could compete if successful.

    Before a substantive pilot begins, agencies should identify success criteria, funding, production assurance requirements, the likely procurement pathway and whether the participating company could compete through that pathway if the pilot succeeds.

  4. Give successful pilots a genuine opportunity to compete, without guaranteeing a contract.

    Success should not guarantee a contract. It should guarantee a genuine opportunity to compete.

  5. Apply an Australian AI Capability Test within existing procurement/architecture/FOCI: before exclusive foreign-AI commitments for strategically significant workloads, consider Australian options, provider/connectivity/jurisdictional concentration, portability/substitutability, Australian continuity and exit cost/feasibility; Australian procurement is not unconditional.

    GovTech Australia proposes an Australian AI Capability Test embedded within existing procurement planning, architecture assessment and FOCI processes. Before committing a strategically significant workload exclusively to foreign-controlled AI capability, agencies should consider Australian capability at the relevant layer, provider/connectivity/jurisdictional concentration, portability and substitutability, the existence of an Australian-based continuity pathway, and the cost and feasibility of exit. This is not a requirement to buy Australian regardless of capability. It is a requirement to look before buying offshore by default.

  6. Establish a Sovereign AI Continuity Requirement: foreign-inference services need executable Australian alternatives proportionate to service-loss consequences; materially AI-dependent functions need tested pathways sustaining defined minimum service through prolonged disruption.

    The Commonwealth should establish a Sovereign AI Continuity Requirement. AI services relying on foreign- controlled inference should maintain an executable Australian-based alternative pathway appropriate to the consequence of service loss. For functions materially dependent on AI, that pathway should be tested and capable of sustaining a defined minimum level of service during a prolonged disruption.

  7. Address connectivity, provider/technology and jurisdictional failures; allow technically viable, periodically tested alternatives, including reduced capability sufficient to sustain government functions.

    The requirement should address three distinct failure domains: • Connectivity: can Australia continue operating if international connectivity is unavailable or materially degraded? • Provider and technology: can the workload continue if a provider, model or compute service becomes unavailable? • Jurisdiction: can Australia continue operating if access is restricted by sanctions, export controls, licensing decisions or another foreign legal measure? • The fallback need not reproduce the primary service exactly. It may use alternative Australian inference, a different model, reserved Australian compute or a deliberately reduced capability sufficient to sustain the government function. A theoretical ability to migrate is not resilience; the alternative must be technically viable and periodically tested.

  8. Require interoperability, portability, substitutability and practical exit, distinguishing geographic, provider, connectivity and jurisdictional redundancy.

    Require interoperability, portability, substitutability and practical exit, distinguishing geographic, provider, connectivity and jurisdictional redundancy.

  9. Measure Australian capability reaching production; where practicable distinguish Australian-controlled technology from expenditure through foreign-owned Australian subsidiaries, capturing Australian SME and subcontractor participation.

    Finally, Government should measure whether Australian capability reaches production. Reporting should distinguish, where practicable, Australian-controlled technology participation from expenditure contracted through Australian subsidiaries of foreign-owned companies, and capture Australian SME and subcontractor participation.

  10. Enable increased Australian control at each layer according to workload sensitivity and consequence.

    Sovereignty is not binary. Government should be able to increase Australian control at each layer according to workload sensitivity and consequence.

  11. Use a federated, contestable architecture rather than a single national platform or champion.

    This should be a federated, contestable architecture, nota single national platform or national champion.

  12. Incorporate continuity and dependency considerations into DTA AI Technical Standard assessments for consequential systems.

    DTA’s AI Technical Standard already addresses architecture, interoperability, lifecycle management and vendor lock-in.[6] For consequential systems, that assessment should incorporate the continuity and dependency considerations outlined above.

  13. Support AI Safety Institute testing and evaluation with enduring technical capability and evaluation environments, including when foreign models or infrastructure are constrained.

    The Australian AI Safety Institute also has a role in strategic resilience. Its ability to test and evaluate emerging systems should be supported by enduring access to sufficient technical capability and evaluation environments, including during periods when access to foreign models or infrastructure is constrained.[8]

Read the original

Open in the interactive guide