Submission 264
Intelligent Screening Australia Pty Ltd (i-screen)
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.
Summary
Read the original submission for the full text.
A small Australian health-technology company presents an operating perspective, arguing Australia should pursue AI adoption and risk management together and supporting clear, proportionate, enforceable regulation covering domestic and offshore suppliers. Covers adoption depth, human oversight, skills, cyber security, sovereign capability and data centres; says it could itself be eligible for some measures it proposes.
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.
Says the findings do not describe the whole economy but indicate concentrated AI capability and demand.
These findings do not describe the whole economy, but they indicate that capability and demand are concentrated.
A program corrected obvious text-recognition errors in this quote for display.
The submission recommends that Australia measure and report depth, completion and applied AI capability rather than incidence alone.
We recommend that Australia measure and report depth, completion and applied capability rather than incidence alone.
A program corrected obvious text-recognition errors in this quote for display.
The submission argues that new AI obligations should be redesigned if equivalent outcomes cannot be enforced across the market, rather than imposed only on reachable domestic firms.
If equivalent outcomes cannot be enforced across the market, the obligation should be redesigned rather than imposed only on readily reachable domestic firms.
A program corrected obvious text-recognition errors in this quote for display.
The submission recommends that Australian standards define effective human oversight by competence, authority to amend, reject or escalate, and accountability.
We recommend that Australian standards define effective oversight through three practical factors: competence relevant to the decision, authority to amend, reject or escalate, and accountability within the organisation for the released outcome.
A program corrected obvious text-recognition errors in this quote for display.
Says an episode it describes shows that access to an externally controlled model can change on short notice for reasons unrelated to an Australian customer's conduct.
The episode shows that access to an externally controlled model can change on short notice for reasons unrelated to an Australian customer's conduct.
A program corrected obvious text-recognition errors in this quote for display.
Says Australia may have a more achievable opportunity in narrow systems grounded in lawfully accessed, well-governed Australian data and domain expertise.
Australia may have a more achievable opportunity in narrow systems grounded in lawfully accessed, well-governed Australian data and domain expertise.
A program corrected obvious text-recognition errors in this quote for display.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Measure introductory AI training registrations, completions, capability gained and workplace application; create pathways into governance and sector-specific implementation.
Australia has announced one million subsidised places in introductory AI training through the National AI Centre and TAFE NSW [6]. The next policy need is to measure registrations, completions, capability gained and workplace application, and to create pathways beyond introductory literacy into process automation, agentic workflows, evaluation, governance and sector-specific implementation.
A program corrected obvious text-recognition errors in this quote for display.
Measure and report depth, completion and applied capability of AI adoption rather than incidence alone.
We recommend that Australia measure and report depth, completion and applied capability rather than incidence alone.
A program corrected obvious text-recognition errors in this quote for display.
Extend national training beyond introductory content into practical process automation, agentic workflows, evaluation, cyber defence and compliance, examining and evaluating international programs before replicating them at scale.
National training should extend beyond introductory content into practical process automation, agentic workflows, evaluation, cyber defence and compliance. International programs should be examined as models, with their outcomes evaluated before being replicated at scale.
A program corrected obvious text-recognition errors in this quote for display.
Consider a voluntary, independently assessed AI assurance mark with transparent criteria, assessor, renewal and complaints processes, explicitly excluding TGA approval, medical-device inclusion or general government endorsement.
voluntary, independently assessed Australian AI assurance mark could support that objective if its criteria, assessor, renewal process and complaints mechanism are transparent. It should state clearly that the mark does not constitute TGA approval, medical-device inclusion or general government endorsement.
A program corrected obvious text-recognition errors in this quote for display.
Before new AI obligations commence, compare comparable jurisdictions, explain material divergence and test enforceability against offshore suppliers; redesign obligations if equivalent market-wide outcomes cannot be enforced.
New AI obligations should therefore undergo two tests before commencement. A competitive-symmetry test should compare the obligation with comparable jurisdictions and explain any material divergence. An enforceability test should identify how the obligation will apply to offshore providers supplying Australian customers. If equivalent outcomes cannot be enforced across the market, the obligation should be redesigned rather than imposed only on readily reachable domestic firms.
A program corrected obvious text-recognition errors in this quote for display.
Prioritise risk-based disclosure of material AI involvement, qualified review and reviewer authority, relevant storage and processing locations, and substantiated capability and assurance claims, without revealing sensitive security or proprietary implementation details.
We recommend prioritising visible and checkable outcomes: disclosure of material AI involvement, disclosure of whether a qualified human reviewed consequential output and what authority they had, meaningful disclosure of storage and processing location where relevant, and substantiation of capability and assurance claims. These disclosures should be risk-based and should not require publication of sensitive security or proprietary implementation details.
A program corrected obvious text-recognition errors in this quote for display.
Define effective human oversight in Australian standards by competence, authority to amend, reject or escalate, and organisational accountability, with the level of oversight reflecting the consequence and reversibility of the decision.
We recommend that Australian standards define effective oversight through three practical factors: competence relevant to the decision, authority to amend, reject or escalate, and accountability within the organisation for the released outcome. The required level of oversight should reflect the consequence and reversibility of the decision rather than applying one control to every use of AI.
A program corrected obvious text-recognition errors in this quote for display.
Teach AI strengths, claim verification, effects of bias and data practices, and when not to rely on outputs, alongside assessment of independent ability.
AI capability should be taught explicitly alongside assessment methods that still establish what a student can do independently. Students should learn where these systems perform well, how to verify claims, how bias and data practices affect results, and when not to rely on an output. The objective should be students who can think independently and use AI competently, not a choice between the two.
A program corrected obvious text-recognition errors in this quote for display.
Evaluate a targeted program supporting Australian-owned businesses to employ senior AI talent in Australia, competitive, sector-neutral, independently assessed and available only against published additionality and public-benefit criteria.
We recommend evaluating a targeted program that supports Australian-owned businesses to employ senior AI talent, in Australia. Any program should be competitive, sector-neutral, independently assessed and available only against published additionality and public-benefit criteria.
A program corrected obvious text-recognition errors in this quote for display.
Direct skills investment towards output validation, evaluation, safety, security and governance as well as model development.
We recommend directing skills investment towards output validation, evaluation, safety, security and governance as well as model development.
A program corrected obvious text-recognition errors in this quote for display.
Publicise standing practitioner channels linking the Office of AI, AI Safety Institute and standards development, with transparent remits and feedback; include small operators alongside vendors, civil society, academia and regulators in sector groups, including health.
We recommend publicising standing practitioner channels connected to the Office of AI, the Australian AI Safety Institute and standards development, with transparent terms of reference and feedback on how contributions were used. Sector working | 14 September 2026 5 197 groups, including health, should include small operators alongside vendors, civil society, academia and regulators.
A program corrected obvious text-recognition errors in this quote for display.
Assess extending or complementing the Small Business Cyber Resilience Service for organisations beyond its generally 19-or-fewer-FTE threshold, excluding owners, given AI-powered cyber attacks.
Australia already funds the Small Business Cyber Resilience Service delivered by IDCARE. It provides prevention support, cyber health checks, advice, incident response and recovery, generally for businesses with 19 or fewer full-time-equivalent employees excluding the owner [18]. Government should assess whether support should be extended or complemented for organisations outside that threshold given the increasing risks associated with AI powered cyber attacks.
A program corrected obvious text-recognition errors in this quote for display.
Assess costed shared or subsidised managed detection and response, prioritising sensitive-data holders and considering accredited private delivery and public capability; certified procurement must preserve each organisation’s residual risk-management responsibility.
A costed feasibility assessment could examine shared or subsidised managed detection and response, prioritising sensitive-data holders and considering accredited private delivery as well as public capability. Procurement of a certified service should not remove the residual responsibility of each organisation to manage its risks.
A program corrected obvious text-recognition errors in this quote for display.
Work with AI providers and accredited security bodies on safe-use pathways for legitimate penetration testing, vulnerability research and defensive analysis, while retaining controls against misuse.
Australia should also work with AI providers and accredited security bodies on safe-222 use pathways for legitimate penetration testing, vulnerability research and defensive analysis, while retaining controls against misuse.
A program corrected obvious text-recognition errors in this quote for display.
For any Australian clinical corpus, require lawful authority, provenance, consent where required, privacy, appropriate de-identification, ethics review, access controls, permitted uses, security, independent oversight and transparent IP/public-benefit allocation; do not assume customer data availability.
Any Australian clinical-corpus initiative should operate under a defined public- interest framework covering lawful authority, provenance, consent where required, privacy protection, appropriate de-identification, ethics review, access controls, permitted uses, security, independent oversight and transparent allocation of resulting | 14 September 2026 6 236 intellectual property and public benefits. Existing customer data should not be assumed to be available for this purpose.
A program corrected obvious text-recognition errors in this quote for display.
Independently assess the size and affordability of domain-model adaptation programs, including specialist talent, secure data preparation, compute, evaluation and domain or clinical validation.
A realistic technical path may involve fine-tuning, distilling or otherwise adapting an existing model rather than training a frontier system from scratch. The principal costs associated can include specialist talent, secure data preparation, compute, evaluation and domain or clinical validation. The size and affordability of such programs should be independently assessed.
A program corrected obvious text-recognition errors in this quote for display.
Condition targeted public support on lawful high-quality domain data, valid desired public research and inability to fund adaptation/validation; require privacy/security, independent evaluation, Australian capability, transparent grants and measurable public benefit beyond private IP.
Targeted public support programs may have leverage where an organisation has demonstrable lawful access to high-quality domain data, can support valid and desired public research but cannot fund adaptation and validation alone. Programs should require clear data rights, privacy and security safeguards, independent evaluation, Australian capability development, transparent grant terms and measurable public benefit. Public funding should not be characterised solely as a mechanism for creating privately controlled intellectual property.
A program corrected obvious text-recognition errors in this quote for display.
Allocate a defined research-support portion to Australian SMEs adapting models with lawful Australian domain data; fund secure preparation and independent validation; assess Australian capability, public benefit, data governance and IP, beyond compute alone.
We recommend allocating a defined portion of research and innovation support to Australian small and medium enterprises adapting models using lawfully accessible Australian domain data. Secure data preparation and independent domain validation should be eligible costs. Assessment should consider useful Australian capability, public benefit, data governance and intellectual-property arrangements rather than compute deployed alone.
A program corrected obvious text-recognition errors in this quote for display.
In open-weight capability-floor feasibility assessment, examine performance, safety, licensing, cost and the people, infrastructure, security and governance required for independent operation and updates.
A more achievable resilience objective may be a capability floor using suitable open- weight models that Australia can independently evaluate, adapt, host and maintain. Foreign-origin open weight models do not remove dependency unless Australia has the people, infrastructure, security and governance to operate and update them independently. Performance, safety, licensing and cost would require formal feasibility assessment.
A program corrected obvious text-recognition errors in this quote for display.
Commission an assessment of Australia's exposure to restrictions or interruptions affecting advanced-model access, identifying affected sectors, substitution options and recovery time.
We recommend commissioning an assessment of Australia's exposure to restrictions or interruptions affecting advanced-model access, identifying affected sectors, substitution options and recovery time.
A program corrected obvious text-recognition errors in this quote for display.
Defer open-weight capability-floor commitment until cost, performance and governance are established; involve the AI Safety Institute in independent evaluation and define implementation and operational ownership.
Government should assess the feasibility of an open-weight capability floor rather than commit to one before cost, performance and governance are established. The Australian AI Safety Institute's existing monitoring and testing remit makes it a natural participant in independent evaluation, while implementation and operational ownership would need to be defined [17].
A program corrected obvious text-recognition errors in this quote for display.
Quantify the scale and location of electricity, land, water, connectivity and approval constraints for domestic model hosting and adaptation.
Domestic model hosting and adaptation require compute, and compute requires electricity, land, water, connectivity and approvals. The scale and location of these constraints should be quantified rather than assumed.
A program corrected obvious text-recognition errors in this quote for display.
Assess the costs and feasibility of sovereign or public-interest compute before direct support or tailored procurement, considering proportionality for smaller projects.
There is also a proportionality issue. Requirements designed for well-capitalised hyperscale facilities may affect a smaller nationally oriented compute project differently. It is plausible that sovereign or public-interest compute would require direct support or a tailored procurement model, but that should follow a costed feasibility assessment rather than an assumption that such a project is affordable or necessary.
A program corrected obvious text-recognition errors in this quote for display.
Separate cost-allocation rules for commercial data-centre operators from planning for actual generation and transmission capacity.
We recommend separating cost-allocation rules for commercial operators, planning for actual generation and transmission capacity, and feasibility assessment of any sovereign compute capability.
A program corrected obvious text-recognition errors in this quote for display.
Independently model regional employment’s scale, duration and location across data-centre construction, operations, energy and supply chains, avoiding guaranteed jobs claims.
Data-centre construction, supporting energy and transmission projects and ongoing operations may create regional employment where suitable land and infrastructure are available. The scale, duration and location of employment should be independently modelled across construction, operations, energy and supply chains rather than presented as a guaranteed regional-jobs outcome.
A program corrected obvious text-recognition errors in this quote for display.
Keep water efficiency and catchment-aware siting central in data-centre policy.
Water efficiency and catchment-aware siting should remain central, and regional employment claims should be supported by transparent lifecycle modelling.
A program corrected obvious text-recognition errors in this quote for display.
Focus regulation on accountability for consequential outputs and verifiable operating outcomes rather than brittle design mandates, including proportionate security.
Regulation should focus on accountability for consequential outputs and verifiable operating outcomes rather than brittle design mandates. Durable requirements include competent human oversight where risk warrants it, meaningful disclosure of material AI involvement and data handling, accurate capability claims, proportionate security and rules that can be applied to offshore as well as domestic suppliers.
A program corrected obvious text-recognition errors in this quote for display.
Evaluate policy against measurable adoption, safety, competition and public-interest outcomes rather than the volume of documentation it creates.
Policy should be evaluated against measurable adoption, safety, competition and public-interest outcomes rather than the volume of documentation it creates.
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.