Submission 165
Social Cyber Institute
Submitter group: Research institutes.
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.
Social Cyber Institute-affiliated authors argue for crisis-mode AI policy, a more detailed national plan, diversification from US technology, binding law, independent impact assessment and an education mission. They seek government funding for an international education collaboration the Institute proposes to lead in Australia. The submission does not represent an agreed view of all Institute researchers.
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 AI policymaking should be treated as a crisis of the highest order rather than as isolated technical or regulatory problems.
Policymaking for AI should be treated as a crisis of the highest order rather than as a series of isolated technical or regulatory problems.
Recommends significantly greater diversification of AI technology and expertise away from the US, self-reliance in key operational sectors and new international partnerships.
Significantly greater diversification of sources of AI technologies and expertise for Australia away from the US must be one of the highest priorities for national policy, with self-reliance in the most important operational sectors as a primary reference point all buttressed by new international partnerships.
Argues that safety institutes evaluate but do not bind, and that Australia's AI Safety Institute appears to let government demonstrate seriousness without binding constraints on the firms it governs.
Safety institutes evaluate; they do not bind. Australia's own AI Safety Institute appears to allow the government to demonstrate seriousness without accepting any binding constraint on the firms it is meant to be governing.
Argues an agency AI use case assessment examines a single system, agency and purpose, and cannot assess cumulative deployment effects or whether deployment should proceed.
A use case assessment examines one system, in one agency, for one purpose, against the AI Ethics Principles. It cannot ask what a class of technology is doing to Australian society, what the cumulative effect of many deployments will be, or whether a deployment should proceed at all.
Says the initial reform budget would need to approach at least $1 billion over five years, with teaching funded at least as much as research.
The initial reform budget would need to approach at least $1 billion over the first five years, with a commitment to education (teaching) at least as great as that allocated to research.
Argues Australia should legislate expectations, require disclosure of major AI providers’ political and lobbying expenditure in Australia, fund independent evaluation, and remove procurement self-attestation.
Australia should therefore legislate the expectations rather than publish them, require disclosure of political and lobbying expenditure by major AI providers operating here, fund evaluation capability that is independent of the firms being evaluated, and remove self- attestation from procurement.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Act urgently and explicitly recognise the AI ecosystem as presenting a national and international crisis.
Australian stakeholders need to act urgently and explicitly recognise the AI ecosystem as presenting a national and international crisis.
Address geopolitical/security risks, AI power and sovereignty, democracy and rights, socioeconomic/environmental costs, and governance failure/institutional lag comprehensively, simultaneously and holistically through national authorities.
We can identify five clusters of urgent action all of which are inter-linked and need to be addressed comprehensively, simultaneously and holistically by national authorities: • geopolitical and security risks • power and sovereignty, who controls AI, and on whose terms • democracy and rights • socioeconomic and environmental costs • governance failure and institutional lag.
Call on every substantive actor (government, peak body or organisation) to adopt a rolling 90-day AI action plan, scaled to the actor, and report publicly at the end of each cycle.
one immediate concrete policy is for every substantive actor with a stake in AI policy to be called upon to adopt a rolling 90-day AI action plan, with the cycle scaled to the actor: government, peak body or individual organisation. They should report publicly against it at the end of each cycle.
Consider an adaptable 90-day-plan template: directive, capability/asset mapping, interim supervising team/helpline, AI policy working group, interim guidelines, literacy training, and later governance and safety evaluation; elements and sequence remain open to debate.
A workable template, proposed in discussion and offered here for debate rather than as a settled model, runs as follows: a top-level directive; capability and asset mapping; an interim supervising team or helpline; progression to a dedicated AI policy working group; interim guidelines pending fuller discovery; AI literacy training; and, in later cycles, development of a governance framework and safety evaluation capability. Not every element will suit every actor, and the sequence is open to argument.
Design a far more specific AI plan, markedly more comprehensive and detailed in its policy ambitions than the current National AI Plan, based on experimentation with the mechanisms of government, business, universities and civil society.
Australia must design a far more specific AI plan that is markedly more comprehensive and much more detailed in its policy ambitions than the current National AI Plan. This plan must be based on experimentation with the mechanisms of government, the country’s business environment, its universities and civil society to meet the circumstances of crisis in AI policy.
Give more radical and high-risk measures greater prominence and transform education at all levels.
More radical and high-risk measures must be more prominent. Education at all levels needs to be transformed.
Consider a Prime Minister-chaired national AI council above policy, infrastructure and assurance; explicitly include AI-enabled threats and agentic systems in critical-infrastructure obligations; publish a delivery dashboard tracking dated national AI roadmap commitments.
Australia could follow Singapore's model of a good AI Plan, directed by a national AI council, chaired by the Prime Minister, sitting above a policy, infrastructure and assurance stack.2 That architecture could take on the EU's more explicit security spine, writing AI-enabled threats and agentic systems explicitly into critical-infrastructure obligations. And Australia could adopt the UK's discipline of a published delivery dashboard, so the national AI roadmap is accountable against dated commitments rather than aspiration and generalities.
Make diversification of AI technology and expertise away from the US a top priority, with self-reliance in the most important operational sectors and new international partnerships.
Significantly greater diversification of sources of AI technologies and expertise for Australia away from the US must be one of the highest priorities for national policy, with self-reliance in the most important operational sectors as a primary reference point all buttressed by new international partnerships.
Establish a public-option capability through public-interest laboratories, lawful high-quality data and reserved domestic compute.
establish a public-option capability through public-interest laboratories, lawful high-quality data and reserved domestic compute
Negotiate dependable frontier access with the United States, subject to review for geopolitical and commercial concentration risk.
negotiate dependable frontier access with the United States, subject to review for geopolitical and commercial concentration risk.
Define sovereignty operationally across model, infrastructure, knowledge, continuity and human-authority layers rather than as technological independence.
Define sovereignty operationally across model, infrastructure, knowledge, continuity and human-authority layers rather than as technological independence.
Look to lead international partnerships, including with Five Eyes members and other like-minded countries.
A key further step for Australia should be to look to lead international partnerships. This can include Five Eyes members but also other like-minded countries.
Provide Australian government funding in support of Australian membership of the International Cyber Education Collaborative.
has proposed that it lead Australian membership of the International Cyber Education Collaborative (ICEC) alongside UK and US partners. Singapore and Canada representatives are also negotiating entry. But funding in support from the Australian government is yet to be forthcoming.
Make the EU, UK, France, Germany, Canada and Singapore first-rank AI partners alongside engagement with the US and China.
The European Union, the United Kingdom, France, Germany, Canada and Singapore should become first-rank AI partners for Australia, alongside engagement with the US and China.
Adopt a layered AI strategy: reserve expensive frontier models for complex research, sophisticated agents and high-stakes tasks, and route routine work to cheaper open-weight or specialised models by cost, capability and risk.
Australia should adopt a layered AI strategy. Expensive frontier models should be reserved for complex research, sophisticated agents and genuinely high-stakes tasks. Routine work should be routed to cheaper open-weight or specialised models according to cost, capability and risk.
Create an open agent identity and accountability standard covering verifiable identity, attributable actions, revocable permissions and auditable delegation.
create an open standard for agent identity and accountability covering verifiable identity, attributable actions, revocable permissions and auditable delegation
Mandate secure-by-design controls including least privilege, sandboxing, offline testing, runtime monitoring and automatic termination of unauthorised behaviour.
mandate secure-by-design controls including least privilege, sandboxing, offline testing, runtime monitoring and automatic termination of unauthorised behaviour;
Accelerate machine-speed defence alongside specialist human capability.
accelerate machine-speed defence alongside specialist human capability;
Recognise AI as an emerging critical-infrastructure sector.
recognise AI as an emerging critical-infrastructure sector;
Regulate data-centre infrastructure, hardware, software administration and data as separate objects with fit-for-purpose controls.
regulate data-centre infrastructure, hardware, software administration and data as separate objects with fit-for-purpose controls.
Establish an independent technology impact assessment capability for AI, sited outside the agencies that deploy AI and reporting to Parliament, with its first assessment commissioned in 2027.
Establish an independent technology impact assessment capability for AI, sited outside the agencies that deploy AI and reporting to the Parliament, with its first assessment commissioned in 2027.
Establish a National AI Education Mission under the National AI Plan, jointly governed by the Commonwealth and states, with scheduled milestone dates and published capability targets.
Establish a National AI Education Mission under the National AI Plan, jointly governed by the Commonwealth and the states, with scheduled milestone dates and published capability targets.
Deliver the National AI Education Mission through nationally recognised units of competency, shared subjects and micro-credentials rather than through degree cycles.
Deliver this plan through nationally recognised units of competency, shared subjects and micro- credentials that can be stood up in months, rather than through degree cycles that will not produce graduates until 2030.
Fund the fast education layer first and let degrees follow; publish numeric, dated ambitions and begin with a national register of AI provision.
Escalation means inverting the order of reform: fund the fast layer first and let the degrees follow. Ambitions should be stated as numbers with dates and published. Begin with a national register of AI provision: we cannot reform what we cannot see.
Possible actions: newly credential Chief AI Officers/SES officers by end-2027; mandate secure-AI/adversarial-machine-learning units in every cybersecurity award by 2028; increase AI policy/law/governance provision tenfold; track commitments on a UK-style public dashboard.
Possible actions could include: • Every Chief AI Officer and SES officer should be newly credentialed by end-2027 • a mandatory secure-AI and adversarial machine learning unit should be included in every cyber security award by 2028 • policy, law and governance provision for AI analysis should be increased tenfold • a public delivery dashboard, based on the UK model, could report against each commitment.
Fund an initial education reform budget approaching at least $1 billion over five years, with teaching funded at least as much as research.
The initial reform budget would need to approach at least $1 billion over the first five years, with a commitment to education (teaching) at least as great as that allocated to research.
Treat the political power and intent of AI corporations as major policy variables, and design regulation so it does not depend on the goodwill, self-restraint or stated principles of any firm.
Treat the political power and intent of the AI corporations as major policy variables in their own right and design Australian regulation so that it does not depend on the goodwill, self-restraint or stated principles of any firm.
Legislate expectations; require disclosure of political and lobbying expenditure by major AI providers operating in Australia; fund evaluation independent of evaluated firms; remove procurement self-attestation.
Australia should therefore legislate the expectations rather than publish them, require disclosure of political and lobbying expenditure by major AI providers operating here, fund evaluation capability that is independent of the firms being evaluated, and remove self- attestation from procurement.
Privilege binding legal obligations over voluntary pledges, holding leading actors to standards of enforceability rather than intent.
Policy should privilege binding legal obligations over voluntary pledges and hold leading actors to standards of enforceability rather than intent.
Immediately study jurisdictions that have converted AI safety positions into binding law (Illinois, California, New York, Minnesota, the EU) as comparators.
Jurisdictions that have converted such positions into binding law (Illinois, California, New York, Minnesota, the European Union) are the appropriate comparators for Australia, and they should be studied immediately.
Accept the procedural step that follows where Australia has accepted an argument in an international forum, as with the 2023 two-tier approach to autonomous weapons.
Where Australia has accepted an argument in an international forum, as it did with the two-tier approach to autonomous weapons in 2023, it should accept the procedural step that follows.
Mobilise effective actors at home and internationally, and secure the political commitments needed to get them into the room.
Build a movement: mobilise effective actors at home and internationally, and secure the political commitments needed to get them into the room.
Secure political commitments to concrete measures such as standing representation for affected populations on AI advisory and governance bodies, funded participation and the same access to officials that industry enjoys.
That commitment should be concrete through measures such as standing representation for affected populations on AI advisory and governance bodies, funded participation rather than unpaid consultation, and the same access to officials that industry currently enjoys.
Resource peak bodies to turn dispersed concern into technically credible submissions and represent workforces and sectors unable to sustain individual engagement.
Peak bodies are the natural aggregators, capable of converting dispersed concern into technically credible submissions, and of representing workforces and sectors that cannot individually sustain engagement. They should be resourced to do so.
Adopt genuine crisis-mode decision-making.
Our prescription follows the same logic point for point: adopt genuine crisis-mode decision-making;
Use technology impact assessment as the methodological foundation for consultation and action.
use technology impact assessment as the methodological foundation for consultation and action;
Act on multiple fronts involving government, corporations, universities, community groups and the military; absent functioning multilateral machinery, be prepared for middle powers to build coalitions rather than wait.
act on a multi-front basis involving government, corporations, universities, community groups and even the military; and, in the absence of functioning multilateral machinery, be prepared for middle powers to build their own coalitions rather than wait for one to arrive.
Treat sovereignty, national security, cyber security and critical-infrastructure resilience as inseparable priorities in pursuing economic advantage.
In pursuing economic advantage, Australia’s AI policy should treat sovereignty, national security, cyber security, and critical infrastructure resilience as inseparable priorities.
Build national AI capability without unmanaged dependence on foreign models, infrastructure or providers, focusing on data residency, model behaviour, inference and auditability.
The central policy task is to build national AI capability without creating unmanaged dependence on foreign models, infrastructure or providers. This means focusing on the main control layers that will enhance sovereignty: data residency, model behaviour, inference, and auditability.