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 198

Kitbag Consulting

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.

Kitbag Consulting, whose related business developed DOLIUM, argues sovereignty depends on suppliers’ legal exposure and agencies retaining operating logic. Proposes procurement tiers, model defaults, governance, exit plans and disclosures, acknowledging costs and that sovereignty does not guarantee security or quality. Annexed responses seek broader AI registers, independent safety oversight, binding data-centre obligations and an AI tax timetable.

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 data residency describes where information is stored, whereas sovereignty describes who can access and withdraw it.

    Data residency describes where information is stored. Sovereignty describes who can access and withdraw it.

  2. Argues that a law passed overseas can compel a supplier regardless of what an Australian contract says or where the data resides.

    A contract signed in Australia cannot override a law passed overseas. A supplier bound by one of these laws must comply with it even when the data resides in an Australian data centre and even when the contract with the Commonwealth says otherwise.

  3. Says seven entities disclosed no AI use, so their silence on later questions differs from a genuine gap.

    Seven entities disclose that they use no AI at all; those are marked with an asterisk, since their silence on the later questions means something different from a genuine gap.

    A program corrected obvious text-recognition errors in this quote for display.

  4. Argues that an agency's operating logic, when configured inside a foreign-controlled platform, leaves with that platform.

    Before an AI capability can act inside an agency, it must be given the agency's own operating logic: its System of Work. When that layer is configured inside a foreign-controlled platform, it leaves with the platform.

  5. Argues that open-source inference models can handle around 98 per cent of organisational work, with frontier models reserved for the roughly two per cent of edge cases.

    Open-source inference models can handle around 98 per cent of the work that generates value inside any organisation. The frontier is for edge cases roughly two per cent of use where the task genuinely cannot be done any other way.

  6. Says its September 2026 review of 73 government AI transparency statements found none disclosing onshore inference, Commonwealth-held weights or foreign legal reach.

    Of 73 government transparency statements reviewed in September 2026, none state whether inference is processed onshore, whether the Commonwealth holds the model weights, or whether any foreign jurisdiction has legal reach over the supplier.

Recommendations

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

  1. Classify AI/cloud sovereignty before purchase by supplier, operating-entity and ultimate-parent domicile and foreign legal exposure, rather than data or infrastructure location.

    Implement a tiered sovereignty classification for AI and cloud procurements. Assess sovereignty based on the legal domicile of the supplier, its operating entity and its ultimate parent company and the foreign jurisdictions to which each may be subject, rather than the location in which data or infrastructure is held. Apply this tiered classification at the point of purchase to establish sovereign characteristics before contractual commitment.

  2. Consider a procurement-rule ownership floor of at least 51% Australian ownership for Tier 1 sovereign classification.

    One option is to give the tiered classification a hard ownership floor in the Commonwealth Procurement Rules, so that a supplier cannot be treated as Australian sovereign at Tier 1 unless it is at least 51 per cent Australian-owned.

  3. Define Tier 1 as Australian-owned, with IP, profits, operating logic, workforce and engineering onshore under Australian law, no foreign jurisdictional exposure and an auditable build chain.

    Tier 1, fully Australian sovereign. Australian-owned, holding intellectual property, profits and operating logic onshore under Australian law, with local workforce and engineering across the full stack, no foreign jurisdictional exposure and an auditable build chain.

  4. Define Tier 2 as Australian-owned and governed with Australian-law accountability, differentiating infrastructure partnership from deeper model or IP integration dependencies.

    Tier 2, Australian-owned with foreign collaboration. Australian-owned and governed, with accountability anchored in Australian law, where a multinational collaboration introduces some dependency. An infrastructure partnership carries materially less risk than deep model or intellectual property integration.

  5. For Tier 3 foreign-parented Australian subsidiaries, condition eligibility on confirmed onshore inference, supply-chain disclosure and a Deed of Sovereignty.

    Tier 3, foreign entity with an Australian subsidiary. A foreign parent whose intellectual property and operating logic remain subject to foreign jurisdiction, delivering onshore benefit through local operations, employment and tax. Eligibility should turn on confirmation that inference is processed onshore, on supply chain disclosure and on a Deed of Sovereignty.

  6. Require exceptional justification and mandatory safeguards for Tier 4 foreign suppliers with no substantive Australian entity, workforce or economic contribution.

    Tier 4, fully foreign with no Australian benefit. No substantive local entity, workforce or economic contribution, with data, logic, intellectual property and profits held offshore. The operational kill switch sits outside Australian jurisdiction. Eligibility should require exceptional justification and mandatory safeguards.

  7. Use sovereignty scoring weighted 60% to sovereign control and 40% to Australian economic contribution, adjusting eligibility thresholds by contract value.

    A scoring table supports the classification, allocating sixty per cent to sovereign control and forty per cent to Australian economic contribution. Absence of foreign jurisdictional exposure carries the heaviest weight at twenty per cent, followed by onshore operating logic at eighteen. Contract value adjusts the threshold, so a supplier acceptable at Tier 2 for a thirty million dollar contract may need Tier 1 safeguards at two hundred million.

  8. Weight freedom from foreign jurisdictional exposure at 20% and onshore operating logic at 18%.

    No foreign jurisdictional exposure | 20% Whether the supplier, its operating entity or its ultimate parent can be compelled under foreign law such as the US CLOUD Act. Physical location is not legal protection. Onshore operating logic 18% Whether the structured logic governing decisions, workflows and authority is held under Australian law rather than as configuration inside a foreign-controlled platform.

  9. Weight Australian ownership at 12% and data/inference localisation at 10%, assessing foreign controlling interests and confirmed onshore inference rather than storage alone.

    Australian ownership 12% Whether a foreign parent holds a controlling interest. A Deed of Sovereignty mitigates this exposure for foreign-parented suppliers but does not replicate Australian ownership. Data and inference localisation 10% Whether inference processing is confirmed onshore, not only data storage. Many platforms route inference through offshore capacity.

  10. Weight Australian workforce and research at 14% and Australian-retained IP at 12%, assessing engineering/research employment and operating logic independent of models or vendors.

    Australian workforce and research | 14% Whether engineers and researchers are employed in Australia rather than a sales presence alone. Skills developed here stay in the Australian economy. Intellectual property retained in | 12% Whether operational ontologies and institutional logic persist as Australia Australian-owned assets, independent of any single model or vendor.

  11. Weight onshore profits at 8% and Australian SME supply-chain participation against defined thresholds at 6%.

    Profits retained onshore 8% Whether economic returns are anchored in Australia rather than flowing to a foreign parent. Australian SME participation 6% Whether the supply chain includes Australian small and medium enterprises against defined thresholds.

  12. Apply the same sovereignty classification to robotics and industrial automation, and update industry and defence policies to recognise domestic robotics integration and manufacturing as sovereign industrial capability.

    Apply the AI sovereignty classification to robotics and industrial automation. Assess robotics and automated production lines using the same jurisdictional criteria applied to AI, establishing the controlling entity’s domicile at the point of purchase. Update industry and defence policies to recognise domestic robotics integration and manufacturing as sovereign industrial capability.

  13. Embed a mandatory sovereignty weighting in value-for-money assessments by updating Commonwealth Procurement Rules evaluation guidance, aligning with New Zealand's mandatory minimum weighting for economic benefit.

    Embed a mandatory sovereignty weighting in value-for-money assessments. Update evaluation guidance under the CPRs to include a sovereignty weighting, aligning with the New Zealand precedent that establishes a mandatory minimum weighting for economic benefit.

  14. Keep sovereignty weightings within Australia’s procurement commitments, including national treatment for covered suppliers above threshold.

    A weighting of this kind must sit within Australia’s procurement commitments. The WTO Agreement on Government Procurement, the Australia United States Free Trade Agreement and comparable instruments each require national treatment for covered suppliers above threshold.

  15. Incorporate a Deed of Sovereignty into Commonwealth AI contracts, requiring supplier and parent foreign-compulsion disclosures, warranties and immediate notification if compulsion occurs.

    Incorporate a Deed of Sovereignty into Commonwealth AI contracts. Require suppliers to identify and warrant every foreign jurisdiction under which they, or their parent entities, may be compelled to produce data, modify a service, or suspend access and to notify the Commonwealth immediately if that compulsion occurs.

  16. Mandate that suppliers notify the Commonwealth of any change to their jurisdictional exposure and provide meaningful business continuity and redress where delivery is disrupted by foreign action, on the Commonwealth's determination of what is essential.

    Mandate that suppliers notify the Commonwealth of any change to this jurisdictional exposure and provide meaningful business continuity and redress where delivery is disrupted by foreign action, on the Commonwealth’s own determination of what is essential to service delivery rather than the suppliers.

  17. Update procurement rules for AI impracticable to build domestically, requiring the strongest available onshore inference, Australian-law operating logic and, where supplier structure permits, Commonwealth-held weights; use Recommendation 6’s alternative otherwise.

    Update the Commonwealth Procurement Rules (CPRs) for imported AI capability. Where AI capability cannot practicably be built domestically, condition its acquisition on the strongest combination of safeguards a supplier can offer onshore inference processing, operating logic retained under Australian law and, where the supplier’s structure allows it, Commonwealth control of the model weights. Proprietary frontier suppliers will rarely offer control of their weights; where that is the case, Recommendation 6 sets out the alternative.

  18. Require major AI investment proposals assessed through the Digital and ICT Investment Oversight Framework to explicitly identify where sovereign control sits.

    Require major AI investment proposals assessed through the Digital and ICT Investment Oversight Framework to explicitly identify where sovereign control sits.

  19. Default to small, open-weight (held and operated onshore) or Australian-developed models for domain-specific capability built on Australian data, rather than frontier models accessed as a service, subject to strict provenance assessments and supply chain disclosures.

    For AI capabilities built on Australian data, agencies should default to small language models that are either open-weight and held and operated onshore, or developed in Australia, rather than defaulting to a frontier model accessed as a service. Subject these deployments to strict provenance assessments and supply chain disclosures regarding the model developer.

  20. Recognise agency operating logic as a sovereign asset, separate from executing systems and staff, retained long term within Australian Government Architecture through a distinct procurement category.

    Hold an agency’s operating logic within the Australian Government Architecture, as a distinct procurement category within existing marketplace arrangements. Recognise an agency’s operating logic as a sovereign asset held separately from the systems and staff that execute it, embedded in the Architecture over the long term rather than inside whichever platform currently executes it.

  21. Require the layer holding an agency's rules, roles, processes and boundaries to be held under Australian law and structurally separated from the underlying AI models.

    Require the layer holding an agency’s rules, roles, processes and boundaries to be held under Australian law and structurally separated from the underlying AI models.

  22. For AI supporting decisions affecting people, require machine-readable permitted-action rules enforced before action, scaled to the existing impact-assessment risk rating rather than a new assessment.

    Require rules-based governance for AI supporting decisions affecting a person, scaled to the risk rating in the agency’s existing impact assessment. For any AI capability supporting such a decision, define what the AI is permitted to do, make those rules machine-readable and enforce them before the point of action, with the rigour of enforcement set by that existing risk rating rather than a new assessment.

  23. Constrain the system’s permitted actions using model outputs, rather than attempting to constrain a model’s internal behaviour.

    It does not require constraining a model’s internal behaviour, which is beyond present capability. It constrains what a system is permitted to do with a model’s output, being which actions, it may take, against which records, under whose authority and within what limits. Those constraints sit outside the model, so they work with a general purpose model an agency did not build and cannot inspect.

  24. Apply defined legislative, delegated or policy rules where authority and process exist; do not encode invented rules for genuinely open judgement.

    The second qualification is scope. Rules of this kind bind well where authority and process are already defined in legislation, delegations or policy, which describes most Commonwealth decision-making. They bind poorly where a judgement is genuinely open and an agency should not pretend otherwise by encoding a rule that does not exist.

  25. Where AI operates across agency boundaries, require each agency to retain authority over its own rules while a shared and inspectable layer lets agents interact within limits every participating agency can see.

    Where AI operates across agency boundaries, require each agency to retain authority over its own rules while a shared and inspectable layer allows agents to interact within limits every participating agency can see.

  26. Require a credible exit path, tested against a defined dependency scenario, for major AI and platform investments before approval.

    Require a credible exit path for major AI and platform investments prior to approval, tested against a defined dependency scenario.

    A program corrected obvious text-recognition errors in this quote for display.

  27. Require agencies to identify failures, duration and recovery if supplier, model or jurisdiction becomes unavailable, plus operating-logic location, substitution requirements and assets retained on exit.

    Mandate that agencies state what would stop working if the supplier, the model or its jurisdiction became unavailable, for how long and what would restore it, alongside where operating logic will be held, what would be required to substitute the capability and what specific assets the agency would retain if the supplier relationship ended.

  28. When suppliers cannot structurally retain specialist skills in Australia, require vendor-neutral training on maintenance, personnel, functionality and resourcing for in-house delivery or re-sourcing, including direct training of Commonwealth staff.

    Where an agency procures AI capability from a supplier that is structurally unable to retain specialist skills within the Australian economy, require vendor-neutral training conditions covering maintenance, personnel, functionality and the resourcing needed to bring the capability in-house or re-source it to another supplier. The vendor should be obliged to train Commonwealth staff directly, not merely document the capability.

    A program corrected obvious text-recognition errors in this quote for display.

  29. For each material AI capability, disclose operating-entity and ultimate-parent domicile, onshore inference status and whether the Commonwealth holds or merely accesses model weights in transparency statements.

    For each material AI capability, the Standard for AI transparency statements should require disclosure of the legal domicile of the operating entity and its ultimate parent, whether inference is processed onshore and whether the Commonwealth holds or merely accesses the model weights.

    A program corrected obvious text-recognition errors in this quote for display.

  30. Have the committee write to each portfolio to classify its current sovereign AI exposure using the tiered classification, including tier changes from supplier ownership or jurisdiction changes, and publish which agencies were able to answer.

    Audit current sovereign AI exposure across portfolios using the tiered sovereignty classification in Recommendation 1. The committee should write to each portfolio to classify its current AI capability against that scale, including where a change in a supplier’s ownership, control or jurisdiction has altered its tier since contract and publish which agencies were able to answer.

  31. Amend the SOCI Act framework to recognise an upstream provider's legal domicile as a hazard a responsible entity cannot mitigate post-contract, and place the obligation to assess it at the point of procurement.

    Amend the Security of Critical Infrastructure (SOCI) Act framework. Formally recognise the legal domicile of an upstream provider as a hazard that a responsible entity cannot mitigate post-contract. Place the regulatory obligation to assess this hazard at the point of procurement.

  32. Treat overt foreign-law compulsion of suppliers as a distinct foreign-interference risk category.

    Foreign interference risk here is not confined to covert activity. The exercise of foreign law over a supplier is overt, lawful in the jurisdiction concerned and generally not disclosable to the Australian customer, which places it outside the frame of most foreign interference assessments. We invite the committee to treat it as a distinct category of risk.

  33. Annex — Chaney consultation: add a sovereign procurement policy and make onshore, Australian-law inference within a governed operational boundary a condition of whole-of-government eligibility for frontier developers.

    We recommend adding Policy 19, as set out below in response to Question 2, and reforming the CCA suitability assessment framework to incorporate a digital sovereignty tier. This would require frontier AI developers to demonstrate that Commonwealth data inference occurs within Australian jurisdiction, under Australian law and within a governed operational boundary, as a condition of whole-of-government contract eligibility.

  34. Annex — Chaney consultation: explicitly pursue sovereign AI exports through National AI Missions, including the Indo-Pacific opportunity.

    The National AI Missions framework should explicitly pursue this opportunity rather than leave it unaddressed. Sovereign AI export is not simply a commercial proposition; it is a foreign policy instrument and a mechanism for extending Australian governance standards and rule-of-law principles into a region where the alternative is a default toward either US platform dependency or Chinese state-adjacent technology. The paper’s silence on this front represents a missed strategic opportunity that compounds with each year the market develops without an Australian position in it.

  35. Annex — Chaney consultation: clarify whether defence AI is intentionally governed separately; if so, say so; otherwise recommend a classified policy track with defined civilian-framework interfaces.

    Whether this gap is intentional, reflecting a deliberate decision to address defence AI through classified policy instruments, or an omission that leaves defence AI development without adequate civilian governance oversight should be made explicit. If the former, the paper should say so. If the latter, a separate classified policy track with defined interface points to the civilian framework should be recommended.

  36. Annex — Chaney consultation: streamline or prefer Tier 1 and Tier 2 sellers, while subjecting Tier 4 sellers to enhanced conditions or exclusion.

    Tier 1 and Tier 2 sellers receive streamlined or preferential assessment. Tier 4 sellers face enhanced conditions or exclusion.

  37. Annex — Chaney consultation: extend the proposed AI register to frontier productivity tools, internal decision support and agentic automation, disclosing systems, developers, sovereignty tiers, inference jurisdiction and deployment layer.

    The register must span all three deployment tiers. It should require disclosure of: the AI system used and its developer; the sovereignty tier under which it is deployed and whether inference occurs onshore under Australian law; the layer of deployment (frontier productivity, internal decision support or agentic automation);

  38. Annex — Chaney consultation: disclose human oversight, its position in the decision chain and the volume of decisions influenced or made.

    what human oversight is in place and at what point in the decision chain it operates; and the volume of decisions influenced or made.

  39. Annex — Chaney consultation: require agencies to demonstrate a governed System of Work providing structural accountability across all three tiers before deploying AI at any layer.

    Agencies should further be required to demonstrate that a governed System of Work providing structural accountability across all three tiers is in place as a condition of deploying AI at any layer.

  40. Annex — Chaney consultation: air-gap frontier productivity models from organisational knowledge.

    The outermost is the frontier model tier, platforms connected to the global AI ecosystem carrying the full jurisdictional and training signal risks described above. These tools must be air-gapped from organisational knowledge.

  41. Annex — Chaney consultation: consider AISI independence through an independent statutory board, fixed-term appointments and published methods; a protected technical core publishing without ministerial clearance; or a statutory ban on ministerial assessment direction.

    The design question that deserves more attention is what governance model protects the AISI’s technical independence. Options worth considering include: • A statutory authority model with an independent board, fixed-term appointments and a published methodology for model assessments, insulating assessments from ministerial direction. • A hybrid structure with a protected technical core that publishes findings without ministerial clearance, alongside a policy advisory function that operates through normal departmental channels. • An explicit provision, modelled on the Australian Bureau of Statistics Act, prohibiting ministerial direction on the content of technical assessments.

  42. Annex — Chaney consultation: enforce frontier transparency obligations at market entry, excluding non-compliant developers from Commonwealth contracts and CCA participation from the outset.

    The transparency obligations in Policy 14 must therefore be enforced as conditions of market entry, not as ongoing reporting requirements that assume a cooperative relationship. A developer that does not meet them should be ineligible for Commonwealth contracts and CCA participation from the outset.

  43. Annex — Chaney consultation: announce binding data-centre obligations and an AI tax framework within six months, complete design within twelve months and enact within twenty-four months.

    The government should publicly announce within six months that binding data centre obligations and an AI tax framework are coming, regardless of whether the design is finalised. The design process should complete within twelve months. The framework should be enacted within twenty-four months.

  44. Annex — Chaney consultation: convert data-centre expectations and voluntary commitments into enforceable terms and a mandatory returns mechanism before infrastructure is built and leverage lost.

    This is precisely the gap that converting the Data Centre Expectations into binding obligations (Policy 17) and establishing a mandatory returns mechanism (Policy 18) are designed to close. The MOU model should be the floor, not the ceiling: a signal of good faith from industry that the government then has an obligation to formalise into enforceable terms before the infrastructure is built and the leverage is gone.

  45. Annex — DTA response: make acceptance of Australian sovereignty terms on jurisdiction, data governance and accountability a non-negotiable eligibility condition.

    Acceptance of Australian government sovereignty terms on jurisdiction, data governance and accountability to Australian law should be a condition of eligibility, not a starting position for negotiation.

  46. Annex — DTA response: apply government terms with minimal adaptation to Tier 1; adapt Tier 2 commercial terms proportionately to collaboration depth while preserving sovereign accountability through a Deed.

    Tier 1 — Australian-owned, subject only to Australian law These sellers face the least friction. Their structures are already aligned with government requirements and government T&Cs should apply with minimal adaptation. Tier 2 — Australian-owned in collaboration with a foreign entity friction may arise where multinational arrangements introduce conflicting obligations. Commercial terms may be adapted proportionately to the nature and depth of the collaboration. A Deed of Sovereignty can help delineate which obligations sit with the Australian entity and which with the foreign partner, preserving sovereign accountability while acknowledging operational reality.

  47. Annex — DTA response: allow Tier 3 commercial carve-outs only for genuine global-structure conflicts, with agency disclosure and approval; exclude jurisdiction, data governance and accountability from carve-outs.

    Tier 3 -- Foreign entity with Australian subsidiaries A defined set of commercial carve-outs may be negotiated where genuine conflicts with global structures exist, subject to disclosure to the relevant agency and approval. These carve-outs apply to commercial mechanics only, not to jurisdiction, data governance, or accountability provisions.

  48. Annex — DTA response: require government terms in full for Tier 4 transnational platforms as a condition of eligibility.

    Tier 4 — Transnational Platforms/Sellers Government T&Cs apply in full as a condition of eligibility. Their jurisdictional exposure is precisely the risk the T&Cs are designed to manage and no commercial carve-out resolves that underlying problem.

  49. Annex — DTA response: embed sovereignty weighting in PGPA and procurement value-for-money evaluation, following the cited New Zealand minimum 10% precedent.

    The PGPA and the Commonwealth Procurement Rules should embed a sovereignty weighting into the value for money evaluation, following New Zealand's precedent of a legislated minimum 10% weighting on sovereign capability.

  50. Annex — DTA response: calibrate CCA seller baseline expectations by sovereignty tier, including Australian investment across data infrastructure, inference, models, applications and governance rather than sales alone.

    The following are reasonable baseline expectations of any CCA seller, calibrated to their sovereignty tier (see scale below): • Investment in Australian operations across the full AI lifecycle, data infrastructure, GPU/inference, foundation models, applications and governance not just sales and account management presence.

  51. Annex — DTA response: require full supply-chain and jurisdictional disclosure identifying where model weights and operating logic are held, foreign entities with legal access and applicable compulsion instruments.

    Full supply chain and jurisdictional disclosure: where model weights and operational logic are held, which foreign entities have legal access and whether the company is subject to instruments such as the US CLOUD Act, which compels US companies to grant access to data regardless of where it is physically stored.

  52. Annex — DTA response: confirm and report Commonwealth data localisation, including onshore GPU inference rather than local storage alone.

    Data localisation for Commonwealth data as a baseline, including GPU inference onshore. Many transnationals route inferences through global AI factories; a user’s request, potentially with attached Commonwealth data, is processed offshore even where storage is local. This must be explicitly confirmed and reported.

  53. Annex — DTA response: separate globally shareable execution infrastructure from agency operating logic, retaining that logic under Australian ownership and jurisdiction.

    Architectural separation between execution infrastructure (which can be globally shared) and operational ontology the logic encoding how an agency functions, complies and exercises authority. That layer must remain under Australian ownership and jurisdiction.

  54. Annex — DTA response: set defined minimum Australian SME subcontracting thresholds as a share of contract value.

    Meaningful Australian SME subcontracting, with defined minimum thresholds as a share of contract value.

  55. Annex — DTA response: require open architecture or interoperability commitments allowing execution-layer substitution without losing institutional knowledge, retaining operating IP as a sovereign asset.

    Open-architecture or interoperability commitments ensuring the government can substitute execution layers without losing institutional knowledge operational IP must accumulate as a sovereign asset, not a platform dependency.

  56. Annex — DTA response: score criteria at 0, 5 or 10 and weight them to 100; consider contract value in determining the sovereignty safeguards required.

    Score each criterion 0 (no / fully foreign), 5 (partial / conditional), or 10 (yes / fully Australian). Multiply each score by its weight to calculate the weighted contribution, then sum all contributions for the final sovereignty score out of 100. Contract value should also be considered: a vendor who qualifies at Tier 2 for a $30M contract may require Tier 1 safeguards at $200M, because the strategic dependency created by a larger arrangement amplifies every jurisdictional risk.

  57. Annex — DTA response: use tier score bands of 75–100, 55–74, 28–54 and 0–27 for Tiers 1–4 respectively.

    Tier guidance: 75-100 = Tier 1 (fully Australian sovereign) | 55—74 = Tier 2 (Australian, foreign entity collaboration) | 28-54 = Tier 3 (foreign entity with Australian subsidiary) | 0-27 = Tier 4 (transnational platforms).

  58. Annex — DTA: Annually publish CCA sellers’ BuyICT reports: spend/supply chains; employment by role, distinguishing engineering/R&D from sales/support; SME contract-value percentages; tier changes; inference location; foreign laws; deeds; Tier 2 collaborations; sovereign-commitment progress.

    Annual reporting by CCA sellers, published on BuyICT, should cover: • Total Australian spend including full supply chain breakdown • Local employment by role type, distinguishing engineering and R&D from sales and support • Australian SME subcontracting volumes as a percentage of contract value • Sovereignty tier classification and any changes during the reporting period • GPU inference path disclosure confirmation that Commonwealth data inference is processed onshore. • Jurisdictional exposure summary: which foreign legal instruments apply to the seller’s data and operational logic • Deed of Sovereignty status where applicable • Nature and extent of any multinational collaboration arrangements (for Tier 2 sellers) • Progress against any sovereign capability commitments made at CCA establishment

  59. Annex — DTA response: introduce a sovereignty maturity indicator distinguishing contractual compliance from active reduction of government dependency on foreign-controlled infrastructure.

    We further recommend the DTA introduce a sovereignty maturity indicator distinguishing sellers who are contractually compliant from those who actively reduce the government’s structural dependency on foreign-controlled digital infrastructure.

  60. Annex — DTA response: recognise Tier 1/2 skill retention in sovereignty weightings; seek disclosure of knowledge location where Tier 2 collaboration reaches IP or model development.

    For Tier 2 sellers, the extent of this dividend depends on the nature of the multinational collaboration. Where the collaboration is at the infrastructure or execution layer, the core skill and knowledge retention benefit remains intact. Where it extends to IP or model development, the DTA should seek disclosure of where that knowledge resides. The DTA should reflect this structural benefit explicitly in the sovereignty tier weighting it is the difference between an AI investment that builds national capability and one that rents it.

  61. Annex — DTA response: condition Tier 3/4 eligibility on free or heavily subsidised, transferable whole-of-government training aligned with SFIA and APS standards, including AI governance, sovereignty assessment, supply-chain audit and foreign-compulsion literacy.

    For sellers who cannot deliver sovereign skill retention by structure, the CCA framework should require explicit labour and skills development commitments as a condition of eligibility, assessed across three horizons: • Short-term: Free or heavily subsidised access to broadly transferable digital skills training, available whole-of-government rather than negotiated agency by agency. Aligned to Skills Framework for the Information Age frameworks (SFIA) and APS Professions standards not product certifications. Must include AI governance literacy: how to evaluate supplier sovereignty claims, audit supply chains across the AI stack layers, assess CLOUD Act exposure and govern deployed models.

  62. Annex — DTA response: require Tier 3/4 sellers to co-design portable, nationally recognised micro-credential pathways with APS Academy, TAFE or universities in cloud architecture, data engineering, cybersecurity, agentic governance and operating-logic design.

    Medium-term: Co-design of nationally recognised, portable micro-credential pathways with the APS Academy, TAFE, or universities in cloud architecture, data engineering, cybersecurity, agentic AI governance and operational logic design. Capability that survives a model deprecation, vendor repricing, or foreign export control decision.

  63. Annex — DTA response: require long-term knowledge transfer from Tier 3/4 sellers through open documentation, interoperability and transition support, retaining agency operating logic and enabling model or execution-layer substitution without knowledge loss.

    Long-term: Structural knowledge transfer: open documentation standards, interoperability commitments and transition support ensuring institutional logic stays with the agency, not the platform. The goal is an APS that can substitute any execution layer or model without loss of operational knowledge. Sovereignty cannot be restored on demand. It must be designed in from the start.

  64. Annex — DTA: Require annual CCA reports on Tier 1/2 headcount growth, roles, R&D and Tier 2 knowledge location; Tier 3/4 vendor-specific/neutral training hours; staff trained by agency/level, accredited qualifications and APS/SFIA mapping.

    We recommend the DTA requires CCA sellers to report annually on: • For Tier 1 and Tier 2 sellers: Australian headcount growth, role types created and R&D investment as evidence of the sovereign skill retention dividend. For Tier 2 sellers, disclosure of where knowledge generated through multinational collaboration resides. • For Tier 3 and Tier 4 sellers: total training hours delivered, split by vendor-specific versus vendor-neutral content the ratio is a direct indicator of whether sellers are building sovereign capability or consolidating ecosystem dependency. • Number of APS staff trained, by agency and role level and nationally accredited qualifications or micro-credentials delivered. • Training content mapped against APS Professions standards and SFIA frameworks.

  65. Annex — DTA response: establish a sovereign System of Work before major ERP or AI transformations, retaining agency logic as the authoritative source for vendor configuration.

    To preserve sovereignty over their operating logic, agencies should have a sovereign System of Work in place before they undertake large technology transformations such as enterprise resource planning deployments and artificial intelligence initiatives. Where the layer exists first, those subsequent procurements draw from and reference the agency’s own logic, and the vendor systems are configured from it and do not become its only record.

  66. Annex — DTA response: recognise Systems of Work in Australian Government Architecture and expect major ERP and AI investments under the Digital and ICT Investment Oversight Framework to demonstrate retained sovereign operating logic.

    We recommend the DTA recognise the System of Work as a defined capability within the Australian Government Architecture, so that major enterprise resource planning and artificial intelligence investments, assessed through the Digital and ICT Investment Oversight Framework, are expected to demonstrate that the agency retains a sovereign operating-logic layer.

  67. Annex — DTA response: establish a System of Work category in the existing Software and ERP Marketplace, giving sovereign options a structured procurement pathway.

    We further recommend the DTA establish a System of Work category within the existing Software and ERP Marketplace, so that sovereign options in this category receive a structured procurement pathway rather than being assessed indirectly, if at all, through procurements scoped around other systems.

Read the original

In some quotes a program corrected obvious text-recognition errors for display.

Open in the interactive guide