The Productivity Commission told Parliament's AI committee it knows of no comprehensive cost-benefit analysis of AI for Australia

Parliament’s Joint Select Committee on Artificial Intelligence has to present its final report by 30 November 2026. Through October its members put questions on notice to the agencies whose numbers anchor the AI debate, and the answers are now on its additional documents page. Read together, several of them say the same thing in different words: the thing asked about has not been measured.

The clearest is from the Productivity Commission (PC), received on 9 November. Senator David Pocock asked how the Commission had weighed AI’s economic benefits against harms that are hard to count, such as discrimination, misinformation, loss of privacy and national security incidents. The PC replied that it had estimated AI’s effect on measured productivity and GDP, and that this is not an economic cost-benefit analysis of AI for Australia. Then it went further.A

We are unaware of any party that has done a comprehensive cost-benefit analysis of AI for Australia.

Productivity Commission, answers to questions on notice, received 9 November 2026

Exhibit A · from the source

What the $116 billion is, and is not

The figure comes from the PC’s December 2025 inquiry report, Harnessing data and digital technology. Box 1.1 judges that multifactor productivity gains above 2.3 per cent are likely over the next decade and calls what follows a back of the envelope estimate: about an extra $116 billion of Gross Domestic Product (GDP) (2023-24 prices) over the next decade. At the low end of the studies it surveyed, the same box puts the gain at about $26 billion.

That is a gross number about output. The PC’s answer says plainly what it leaves out: Empirically, the PC has not estimated the economic benefits from generative AI. It does not include the harms the question listed, such as discrimination, privacy loss and security incidents, and the PC notes that productivity growth is not a complete measure of economic welfare.

Three more of Senator Pocock’s questions drew the same kind of answer.

The workplace gap is the widest

Senator Tony Sheldon put the consultation question to the Department of Employment and Workplace Relations (DEWR) at the committee’s 30 October hearing. Employers already have consultation duties under work health and safety law, the Fair Work Act, awards and enterprise agreements, and he asked whether anyone knows if they are being met when AI arrives. DEWR’s written answers say that neither it, the Fair Work Commission nor Safe Work Australia holds that data.

• DEWR itself: The department does not have data on the number of disputes that may be raised at the workplace level.
• The Fair Work Commission: There is no requirement for the FWC to record in administrative data whether a given matter included AI as an issue. The FWC told DEWR such issues may be discussed in individual FWC Decisions, which are published one by one.
• Safe Work Australia: its current data collections do not capture information on employer consultation activities, including consultation on WHS matters when AI is being introduced into a workplace.

What DEWR offers instead is guidance. The answer points to the Minister for Employment and Workplace Relations’ announcement on 18 September 2026 that the government would develop best practice guidance on AI adoption through the tripartite AI Employment and Workplaces Forum, which will include evidence and case studies. It gives no date for it.

Two answers that withheld rather than lacked

Not every answer was an absence. The Attorney-General’s Department confirmed it has sought legal advice on copyright and AI, including constitutional law issues, and, citing longstanding practice, declined to say what that advice says. The Department of the Prime Minister and Cabinet, asked which Australian companies train AI models overseas, answered in one line: Information on where Australian companies undertake AI model training is commercial-in-confidence.

And when the Department of Industry, Science and Resources was asked for the evidence behind the AI Safety Institute’s warning of red flags about open-weight models, which we reported on 7 November, its answer offered A sample of public reporting from October: four items from outside organisations. The answer lists no evaluation by the Institute itself. We have not yet read the four items, so we say nothing here about what they found.

Our view

The PC deserves credit for the candour. It answered the narrower question it was asked in 2025, said so, and did not let its $116 billion stand in for a verdict it never reached. The problem is how that number travels. Senator Pocock’s own question opened: You’ve said AI could add more than $116 billion to Australian economic activity over the following decade. That framing treats a gross output figure as the case for AI, while the PC says nobody, as far as it knows, has done the sum with the costs in it.

The workplace answers matter more to most readers, because they describe the present rather than a projection. On these answers, if an employer rolls out algorithmic rostering or monitoring without consulting staff, neither DEWR, the FWC nor Safe Work Australia would know, unless a complaint reached a state safety regulator or a dispute happened to reach a published FWC decision. The PC’s own 2025 report told the government that where gap analyses of the law had not begun they should begin immediately. The same logic applies to the data. Recording whether a Fair Work matter involves AI is a field on a form, not a research program. A committee that reports by 30 November on AI and work could ask for that before it asks for anything harder, because without it every later argument about AI and Australian workplaces will be made on anecdote.