Executive Summary
On 15 July 2026, Prime Minister Anthony Albanese committed Australia to a whole-of-government Artificial Intelligence framework that will include mandatory provisions. An Office of AI was created inside the Department of the Prime Minister and Cabinet, binding national standards will go to Parliament in early 2027, large data centres will be required to pay their own way on power and water, and a copyright exception for training AI on Australian creative work has been ruled out. Voluntary guidance will be superseded by legislated national standards. The government deserves credit for making progress regarding the transformative impact of AI.
An institutional and regulatory advance, however, is not a national transition strategy. The July announcement provides no new funding, no universal framework for worker transition, no sufficiently scaled skills entitlement and no new mechanism for coordinated research or balanced regional adoption. The success of the Hawke-Keating economic reforms was achieved by creating a consensus that brought together government, business, unions, institutions and community groups. AI demands an equivalent shared purpose if employers are to fund training, workers to accept and shape change, educators to rebuild curricula and businesses to co-invest in research.
OCNUS therefore strongly recommends a National AI Summit, convened as soon as practicable and well before the 2027 bill is drafted, to negotiate five funded, time-bound and publicly accountable accords. This is needed so that all sectors contribute to government policy and bodies outside government can commit to a common purpose.
What the Albanese Plan Delivers: Credit Where It Is Due
Almost exactly a year before Albanese’s announcement, OCNUS Consulting called for a National AI Summit. The 2026 plan is not that summit. But set against the gaps in Australian AI policy, it is a good step forward. The Office of AI begins work immediately, charged with coordinating work on energy, copyright, safety, employment, education and security, areas where responsibility was previously scattered across regulators and departments. Together with the AI Safety Institute announced last year, it gives a once-fragmented institutional landscape a centre.
The stated legislative timeframe, aiming for early 2027, has value in itself. Regulatory uncertainty discourages investment and forward planning in every sector. A commitment to binding standards, though the details are still to be settled, at least lets business, unions and other bodies plan against a fixed date.
Energy and water are legitimate public concerns. The announced conditions will oblige large data centres to pay their own way on power and water. Other countries have socialised these costs and come to regret it. Pricing them in from the outset is one of the best elements of the Albanese package.
Finally, the copyright baseline is a genuine point of Australian difference. Few governments have stated their position so plainly: Australia has set itself against a broad commercial text and data mining exception more explicitly than several comparable jurisdictions, although the legal mechanism remains to be designed. It will be a challenging area to navigate, but the stance deserves support.
What the Albanese Plan Omits: Four Gaps
The first omission concerns workers. Existing consultation and safety duties provide partial protection, but Australia has no universal, enforceable entitlement to advance notice, meaningful consultation, retraining, redeployment support or income protection before AI materially changes or eliminates a job, and nothing in the July plan creates one. And because trade union membership has dwindled, no single peak body can now represent the whole workforce, so these protections must come from government. Australian workers deserve nothing less.
Skills are the second gap. The announcement does nothing meaningful to lift AI skills, from the classroom to professional development. What exists is fragmented, modest relative to the likely scale of occupational change and unsupported by any national entitlement to paid, accredited and portable AI training. The government’s main offer to date, a million subsidised places in a short online course, is insufficient: only about a quarter of Australians have undertaken any AI-related training.
Money is the third omission. The 2025 National AI Plan identified more than $460 million in existing commitments, alongside research, commercialisation and AI Accelerator measures, but the July 2026 announcement added no new appropriation. Nor did it create any mechanism for coordinated, mission-based research co-investment to draw business capital and worker savings into partnership with universities and the CSIRO. And $460 million is a paltry sum beside the more than $100 billion in private data-centre investment the government’s own plan anticipates.
The adoption divide is the fourth gap. AI uptake in Australia varies widely by firm size, sector and region, and the plan does not address these divides. Existing programmes recognise the problem, but their scale, reach and accountability are not yet proportionate to it. Some features of the plan may even widen them: the government intends to fast-track AI investment, which flows first to those already positioned to absorb it, and mandatory standards impose compliance costs that large firms carry more easily than small ones. That is a recipe for concentrated adoption, which would deepen urban-regional inequality and forfeit the gains that flow when a new technology spreads across the whole economy.
The scope of the Albanese plan is broad, covering education, training, workplace rights and productivity alongside energy and copyright. That is progress in governing how AI may be used. But standards cannot fund training or create an entitlement. As with the 1980s economic restructuring, Australia needs more than new regulations to succeed in the AI transition.
The Need for a National Summit: Turning Ambition into Strategy
The existing consultative machinery is not equal to the task. The AI Employment and Workplaces Forum is valuable, but a standing forum gathers views where a summit extracts reciprocal commitments. And the Forum’s mandate and representation are too narrow to negotiate across skills funding, regional adoption, research infrastructure, copyright and public-sector implementation. A National AI Summit would not be an
alternative to the Albanese plan. It would be a means of achieving its ambitions. A government can create an office, set a legislative date and attach conditions to data centres. But this will not be enough to incentivise employers to invest in retraining, encourage unions to support AI deployment, help educators rebuild curricula or motivate businesses to put capital into research centres. At least, not enough to transform the nation.
Consensus must be scaffolded by accords that address five areas:
A national AI skills accord establishing a tripartite AI Skills Transition Fund, financed jointly by government and employers, its training designed by unions, educational institutions and industry, delivered in paid working time and carrying portable, nationally recognised credentials. The fund would build on the National Skills Agreement, TAFE and the Jobs and Skills Councils, not beside them.
A workplace transition agreement committing the government to legislate rights to consultation and bargaining where AI materially changes jobs or makes consequential decisions about workers, and employers to provide plain-language impact statements, safety assessments and retraining. These rights would supplement the Fair Work Act and existing safety duties, not duplicate them.
An SME and regional adoption compact closing the gap between large and small enterprises and between the cities and the regions, backed by state and federal co-funding, with support conditional on safe use and workforce engagement. The compact would extend AI Adopt and the National AI Centre into the firms and regions they do not yet reach.
A research co-investment pact beginning with a government pledge of new money, then drawing business capital and, on commercial terms, the worker savings held in industry superannuation into research partnerships with universities and the CSIRO. The last Accord built that pool of savings. This one would put it to work.
An agreement between creators and AI developers establishing an adequately funded collective licensing mechanism, through which developers obtain consent to train on Australian creative work, and creators are paid for its use, with transparent records of the material used in training, an independent umpire for disputes and specific protection for Indigenous Cultural and Intellectual Property. The exchange of consent for payment gives the plan’s copyright baseline its working parts.
Australia has done this type of thing before. In 1983, the Hawke government brought business, unions, the states and community groups to the same table and negotiated commitments until a strategy was established. AI requires the same: a central compact surrounded by sector-specific machinery. Without it, not only will AI diffuse unevenly across society and the economy, but Australia will be left behind.
The summit must be genuinely national, taking in large and small business, unions and non-unionised labour, the states and territories, educational institutions, research agencies, creative industry bodies and community and consumer voices. The Office of AI should serve as its secretariat and afterwards as administrator of the accords, and administering them is what would give it the speed, expertise, authority and adaptability the task demands.
Conclusion: OCNUS Calls on the Government to Convene a National AI Summit
The Albanese plan has earned the credit this paper gives it. Voluntary guidance is moving towards mandatory standards, a scattered policy landscape now has a centre and the copyright baseline marks a genuine point of difference. But a framework that governs the technology is not a strategy for the people and the organisations that must adapt to it. The plan offers workers no enforceable entitlements, adds little to the skills system, commits no new money and leaves the adoption divide to widen. No office, legislative date or condition on a data centre can supply what those gaps demand: employers who fund retraining, workers who shape change, educators who rebuild curricula and businesses that co-invest in research.
The Hawke-Keating lesson still holds, but not as a template. AI is a fast-moving, internationally supplied technology cutting across federal and state responsibilities, and the institutional setting is materially different from the centralised wage-setting of the 1980s. Nor can one summit resolve a national transformation. The transferable lesson is that durable reform is more likely when government establishes a shared diagnosis, negotiates reciprocal commitments and creates institutions capable of sustaining them. Bring all parties into one room, agree on a nationwide commitment, share the risks and spread the rewards. The five accord proposals in this paper are vital for turning Albanese’s AI plan into a national strategy.
The window is short. Consensus must be built before the 2027 bill is drafted, not argued over afterwards. OCNUS therefore calls on the Prime Minister to convene a National AI Summit without delay.