Submission 91
Australian Artificial Intelligence Society Ltd
Submitter group: Industry bodies.
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.
Proposes a 24-month 'Build-Diffuse-Assure' implementation pilot that upgrades existing national AI adoption infrastructure rather than creating new bodies. It covers workflow-first SME support, sector testbeds, a capability framework, interoperable assurance evidence, targeted Commonwealth transparency, and licensing, provenance and Indigenous data governance.
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 policy should track adoption maturity rather than a single usage rate, noting that ABS and NAIC measures describe different layers of adoption.
The ABS reports 12% of all businesses used AI in 2024-25 and explicitly cautions that its measure does not capture intensity or extent of use. NAIC’s SME tracker reports much higher adoption in its surveyed SME population. Policy should therefore track adoption maturity, not a single binary “uses AI” rate.
A program corrected obvious text-recognition errors in this quote for display.
Argues that public policy should reward verified organisational value and capability rather than tool purchases or headline adoption counts.
Public policy should therefore reward verified organisational value and capability—not the purchase of AI tools or headline adoption counts.
A program corrected obvious text-recognition errors in this quote for display.
Proposes a workflow-first protocol under which existing technology and 'no technology change' are considered before custom AI.
Technology selection follows the problem and baseline. Existing functionality, configuration, conventional automation, and “no technology change” must be considered before custom AI.
A program corrected obvious text-recognition errors in this quote for display.
Defines sovereign capability as the ability to understand, adapt and operate critical capability, not a requirement to build everything domestically.
Sovereign capability is the ability to understand, adapt, verify, operate and substitute critical capability under disruption—not a requirement to build everything domestically.
Recommends that the AI Safety Institute focus on advanced risks, with ordinary deployment assurance distributed across organisations and regulators.
The AISI should remain focused on advanced capabilities, risks and harms; ordinary deployment assurance should be distributed across organisations, professions, auditors and regulators.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Recommend a bounded 24-month implementation pilot using existing institutions where possible, with independent evaluation and an explicit scale, modify or stop gate.
recommend a bounded 24-month implementation pilot, using existing institutions wherever possible, with independent evaluation and an explicit SCALE / MODIFY / STOP gate.
Track adoption maturity rather than a single binary AI usage rate.
Policy should therefore track adoption maturity, not a single binary “uses AI” rate.
A program corrected obvious text-recognition errors in this quote for display.
Reward verified organisational value and capability rather than AI tool purchases or headline adoption counts.
Public policy should therefore reward verified organisational value and capability—not the purchase of AI tools or headline adoption counts.
A program corrected obvious text-recognition errors in this quote for display.
Upgrade NAIC and AI Adopt with a common National AI Implementation Protocol that is workflow-first, baseline-led and outcome-measured, rather than creating a new network.
Upgrade NAIC / AI Adopt with a common National AI Implementation Protocol—workflow-first, baseline-led and outcome-measured—rather than creating a new network.
A program corrected obvious text-recognition errors in this quote for display.
Require the protocol to test existing software, configuration and conventional automation first, accepting no AI or no technology change as a legitimate outcome.
The protocol must test existing software/configuration/conventional automation first; “no AI / no technology change” is a legitimate outcome.
A program corrected obvious text-recognition errors in this quote for display.
Pilot up to 1,000 SMEs/family businesses with regional and low-adoption-sector participation; cap indicative Commonwealth support at $10,000–$15,000 per workflow, subject to market sounding, design, co-contributions and appropriations.
Pilot a first cohort of up to 1,000 SMEs/family businesses, with explicit regional participation and coverage of low- adoption sectors. The unit of support is one measurable workflow—not generic digital-transformation consulting. An indicative Commonwealth contribution ceiling of $10,000-$15,000 per workflow would cap direct pilot support at approximately $10-$15 million for a 1,000-business cohort, subject to market sounding, program design, co-contribution settings and available appropriations.
Require the workflow protocol to identify a human owner, measure baseline outcomes, consider existing software and non-AI alternatives, review security and contractual constraints, pilot with oversight and stop conditions, and scale only on evidence.
Define one costly, repetitive, information-heavy, or safety-relevant workflow and its human owner. MEASURE Record baseline time/cost/quality/error/service/risk indicators before choosing technology. ALTERNATIVES Test existing software, configuration, process redesign and conventional automation before custom AI. SECURE Review data, privacy, identity/access, logging, provider/model choice and contractual constraints. PILOT Run a bounded pilot with human oversight and stop conditions. Compare with baseline; include quality, workforce and risk—not only time saved. Scale only when evidence supports it; otherwise modify, stop or retain the existing process.
A program corrected obvious text-recognition errors in this quote for display.
Require each publicly supported pilot to contribute a de-identified Outcome Ledger record of its problem, baseline, intervention, controls, results, failure modes and lessons.
AI Outcome Ledger. Each publicly supported pilot contributes a de-identified evidence record: problem, baseline, intervention, controls, result, failure modes and lessons.
A program corrected obvious text-recognition errors in this quote for display.
Use NAIC for coordination, AI Adopt Centres and competitively selected partners for delivery, education/professional bodies for capability, AISI for advanced-risk science and existing regulators for law and enforcement.
Use NAIC for national guidance and coordination; AI Adopt Centres and competitively selected partners for delivery; universities, TAFEs and professional bodies for capability uplift; the Australian AI Safety Institute for advanced-risk science; and existing regulators for law and enforcement.
A program corrected obvious text-recognition errors in this quote for display.
Create new entities only when an existing institution demonstrably cannot perform the function.
New entities should be created only where an existing institution demonstrably cannot perform the function.
Create competitively selected sector AI testbeds and data partnerships only where Australian data, expertise, regulatory settings or operating conditions create a genuine advantage.
Create competitively selected testbeds and data partnerships only where Australian data, expertise, regulatory settings or operating conditions create a genuine advantage.
Combine lawful datasets, rights/provenance metadata and privacy controls in competitively selected data partnerships where Australia has a genuine advantage, with CARE-aligned governance and Indigenous authority for Indigenous data.
Create competitively selected testbeds and data partnerships only where Australian data, expertise, regulatory settings or operating conditions create a genuine advantage. They should combine: e lawfully curated datasets with rights/provenance metadata, privacy controls and—where Indigenous data are involved—CARE-aligned governance and Indigenous authority over relevant data;
Use sector-specific retrieval, smaller/domain models or adaptation only where evidence shows advantages over generic alternatives in accuracy, cost, latency, privacy, resilience or control.
sector-specific retrieval, smaller/domain models or adaptation methods only where evidence shows advantages over generic alternatives in accuracy, cost, latency, privacy, resilience or control;
Provide secure test environments, evaluation datasets and benchmark suites that allow comparison without unnecessarily exposing sensitive operational data.
secure test environments, evaluation datasets and benchmark suites that allow systems to be compared without unnecessarily exposing sensitive operational data;
Provide proportionate access to compute, evaluation infrastructure and specialist capability for SMEs, researchers and not-for-profits.
proportionate access to compute, evaluation infrastructure and specialist capability for SMEs, researchers and not-for-profits; and
Create commercialisation and operationalisation links between Australian researchers, local suppliers and validated industry problems.
commercialisation and operationalisation links between Australian researchers, local suppliers and validated industry problems.
Define Foundation, Applied and Assurance/Leadership capability bands with sector modules and evidence of applied competence; map them into existing VET, higher-education, microcredential and professional-development systems rather than a parallel passport.
Define three nationally recognisable capability bands—Foundation, Applied, Assurance/Leadership—with sector modules and evidence of applied competence. Map them into existing VET, higher-education, microcredential and professional-development systems rather than creating a parallel “passport”.
Supported pilots should document worker consultation, task redesign, human-approval points, training needs and job-quality effects.
Supported pilots should document worker consultation, task redesign, human-approval points, training needs and job-quality effects.
Adopt a common AI Evidence Passport that crosswalks to existing Australian and international assurance frameworks, with the AISI kept focused on advanced capabilities and ordinary deployment assurance distributed across organisations, professions, auditors and regulators.
Australia already has a national framework for AI assurance in government, NAIC implementation guidance and an Australian adoption of ISO/IEC 42001. AUAI therefore recommends a common AI Evidence Passport that crosswalks to those instruments, NIST AI RMF/GenAl guidance and tested international approaches such as Singapore’s AI Verify. The AISI should remain focused on advanced capabilities, risks and harms; ordinary deployment assurance should be distributed across organisations, professions, auditors and regulators.
A program corrected obvious text-recognition errors in this quote for display.
Include intended use, provenance, evaluation, human oversight, security, monitoring and incidents in a reusable deployment evidence record, crosswalking to existing Australian and international assurance frameworks.
A reusable deployment evidence record containing intended use, data/provenance, evaluation, human oversight, security, monitoring and incidents. It should crosswalk to existing Australian and international assurance frameworks so evidence can be reused across procurement, boards, insurers and regulators.
Require independent assurance or certification only where justified by risk, law, procurement or sector rules; do not treat the Evidence Passport as approval, certification or a safety guarantee.
Important: the AI Evidence Passport is not government approval, certification or a guarantee of safety. It is an interoperable evidence container. Independent assurance or certification should be required only where risk, law, procurement or sector rules justify it.
A program corrected obvious text-recognition errors in this quote for display.
Record accountable executives, affected groups and prohibited uses; data sources, licensing and sensitive information; evaluation benchmarks, limitations, failure modes and human/non-AI comparisons.
Purpose & owner Intended use; accountable executive; affected groups; prohibited uses Data & rights Sources; provenance; permissions/licensing; personaUsensitive information Evaluation Task benchmarks; limitations; failure modes; comparison with human/non-AI baseline
A program corrected obvious text-recognition errors in this quote for display.
Record review, access, security, logging, escalation and model/provider change controls, plus drift, complaints, incidents, periodic review and retirement criteria.
Controls Human review; access; security; logging; escalation; model/provider change control Monitoring Performance/drift; complaints; incidents; periodic review; retirement criteria
For consequential or high-impact Commonwealth AI use cases, publish a standard record (purpose, accountable owner, affected groups, impact-assessment summary, human review, monitoring/incidents, contestability), with explicit, lawful and proportionate national-security, law-enforcement, privacy and operational exemptions.
for consequentiaVhigh-impact use cases, publish a standard record of purpose, accountable owner, affected groups, impact-assessment summary, human_ review, monitoring/incidents and contestability. National-security, law-enforcement, privacy and operational exemptions should be explicit, lawful and proportionate.
Support collective/direct licensing, interoperable provenance/authenticity metadata, transparent dataset governance and proportionate deepfake measures; for Indigenous data, knowledge or cultural material, apply CARE and Australian governance practice, including community authority and benefit-sharing.
Government should support workable collective/direct licensing, interoperable provenance and authenticity metadata, transparent dataset governance and proportionate deepfake/content-authenticity measures. Where Aboriginal and Torres Strait Islander data, knowledge or cultural material are involved, implementation should apply the CARE Principles—Collective Benefit, Authority to Control, Responsibility and Ethics—and draw on Australian Indigenous Data Network / ARDC governance practice, including appropriate community authority and benefit-sharing.
Publish evidence and scale only what works, using existing institutions first.
Use existing institutions first; publish evidence; scale only what works.
Report publicly on productivity, quality and cost, and adoption maturity separately by region, firm size and sector.
baseline-to-result productivity/service improvement, quality/error effects and implementation cost; © adoption maturity: experimentation > bounded workflow > integrated operational use > assured/monitored use, reported separately by region, firm size and sector;
Report worker capability, consultation, human accountability, job redesign and quality; privacy/security incidents, near misses, complaints and remediation; existing-technology versus custom-AI solutions; and no-AI recommendations.
worker capability, consultation, human-accountability design, job redesign and job-quality effects; © privacy/security incidents, near misses, complaints and remediation; e share of problems solved by existing/configured/integrated technology versus custom AI builds; e share of assessments recommending “no AI / no technology change” (an anti-solutionism and value-for-money integrity metric);
A program corrected obvious text-recognition errors in this quote for display.
Report Australian research/supplier participation, sector test assets, reusable evidence and later adoption time/cost savings from Outcome Ledger and Evidence Passport reuse.
Australian research/supplier participation, sector test assets and reusable evidence created; and e reduction in time/cost of later adoption when Outcome Ledger and Evidence Passport evidence is reused.
Independently evaluate the pilot against productivity, inclusion, workforce, security and value for money, with an explicit ability to stop ineffective mechanisms.
The pilot should be independently evaluated against productivity, inclusion, workforce, security and value-for-money outcomes, with an explicit ability to stop what does not work.
Procure delivery through open competition, with bounded co-investment, independent evaluation, international interoperability and an ability to stop mechanisms lacking demonstrated value.
We recommend an operational upgrade to existing infrastructure, open competition where delivery is procured, bounded co-investment, independent evaluation, international interoperability and a genuine ability to stop mechanisms that do not demonstrate value.
Assess whether each mechanism adds value over existing NAIC/AI Adopt support.
whether each mechanism adds value over existing NAIC/AI Adopt support.
During deployment, generate Outcome Ledger records and Evidence Passports and monitor cost, quality, workforce, security and inclusion outcomes quarterly.
Deploy bounded SME/regional cohorts and selected Quarterly de-identified evidence; first sector sector testbeds; generate Outcome Ledger records and : . we 6-18 months F . benchmarks; interim MODIFY / STOP decisions AI Evidence Passports; monitor cost, quality, workforce, . . : where needed. security and inclusion outcomes quarterly.
A program corrected obvious text-recognition errors in this quote for display.
Read the original
In some quotes a program corrected obvious text-recognition errors for display.