Seventy-eight percent of respondents in an Australian poll preferred keeping permission requirements for creative work used in AI training, even if some companies chose not to develop models locally. Nine percent preferred changing the law to encourage more domestic development. The result comes from a RedBridge Group survey of 1,524 adults commissioned by the Australian Recording Industry Association and released on 26 October.[1]
The recording industry has a direct interest in the licensing debate, and the question presented a specific trade-off. The full report adds useful context: a direct question about changing copyright law drew 56 percent opposition and 22 percent support. A subsequent investment-condition question drew 55 percent opposition and 20 percent support. The 78-percent figure comes from a different, forced-choice framing that also allowed an unsure response.[2]
The wording changes what the number means
Three copyright questions in the same survey
| Question format | Preserve permission / oppose change | Change law / support change |
|---|---|---|
| Direct support or opposition to changing the law | 56% | 22% |
| Change required for a company to train locally | 55% | 20% |
| Choose permission or more domestic AI development | 78% | 9% |
Support for local AI development did not translate neatly into opposition to licensing. Of the 33 percent who supported AI companies developing and training models in Australia, 82 percent favoured creators deciding whether their work was used, and 71 percent chose to retain permission in the forced-choice question. On the direct law-change question, however, that subgroup was more divided: 47 percent supported change and 34 percent opposed it.[2]
RedBridge recruited Australian citizens aged 18 and over from opt-in online panels between 8 and 10 October. It weighted responses for demographic and socioeconomic characteristics. Weighting reduced the effective sample size to 884; the report gives an indicative equivalent margin of error of about 3.3 percentage points, while warning that this does not capture coverage or self-selection error.[2]
A disclosed sequence, with a limit on inference
The report publishes the questions in their survey order and cautions that later answers should be read in the context of earlier information. That is a concrete reason to examine the sequence before treating the largest percentage as a context-free measure of opinion. It does not show how respondents would have answered a differently ordered or independently framed survey.[2]
The policy signal
Many respondents wanted creators to retain a say even when local investment was presented as a possible cost. That preference is consistent across the permission questions, though its measured strength changes with the framing. It establishes a political constraint for licensing proposals without determining their price or contract terms.[1]
Sources & further reading
- Four in five Australians say don't change copyright laws for big tech Australian Recording Industry Association · 26 Oct 2026 · Accessed 26 Oct 2026
- Australian AI Poll: AI development, copyright and creative work RedBridge Group / ARIA · Accessed 26 Oct 2026