Submission 154
Dr Ross Howden
Submitter group: Individuals.
Position labels are not shown for individual submitters.
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.
Argues that screen productions cannot resolve opaque upstream training rights: model providers should substantiate lawful training, while productions retain responsibility for their own inputs, authorship, consent and incorporated material. Proposes targeted clarification and tested assurance, with extra evidence only for demonstrated gaps in existing processes. Declares a related patent application, without seeking product endorsement or procurement.
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 use alone does not justify a new verification system, given existing contracts, warranties and chain-of-title practices.
AI use does not, by itself, justify a new verification system. Screen production already relies on contracts, producer warranties, chain-of-title material, performer consent and funding declarations.
Says productions may lack practical visibility of training works, licences and permissions and cannot prove facts about a vendor's corpus they cannot inspect.
Where copyrighted creative material has been used to train a model, the downstream production may have no practical visibility of the source works, licences or permissions involved, even though output from that model can enter a new commercially exploited work. A production cannot prove facts about a vendor's training corpus that it cannot inspect.
Argues that responsibility should sit with the party holding the evidence: providers for training, productions for what they supply and incorporate.
Model providers should bear primary responsibility for substantiating the lawful basis on which protected material was used in training. Productions should remain responsible for material they themselves supply to models and for the authorship, consent, permissions and chain of title of material they choose to incorporate into the work.
States that Australian law offers no settled position on the human authorship needed for copyright in AI-assisted work.
Australian law offers no settled position on how much human involvement is required for copyright in AI-assisted work
Argues that synthetic voice and digital doubles should sit within clearly expressed performer consent and contractual permissions.
Synthetic voice, digital doubles and other generated performances should sit within clearly expressed performer consent and contractual permissions
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Keep existing production mechanisms as default; avoid a parallel AI-specific evidence regime merely because declarations could be false; require extra evidence only for demonstrated inadequacies in authorship, ownership, consent or chain-of-title processes.
Keep existing production mechanisms as the default. Do not create a parallel AI-specific evidence regime merely because a declaration could be false. Require additional evidence only where a specific inadequacy in existing authorship, ownership, consent or chain-of-title processes has been demonstrated.
Where protected material trains a model offered into Australian creative production, place primary lawful-use accountability with its provider; consider the transparency or auditable evidence needed to substantiate it.
Place training-corpus accountability with model providers. Where protected material is used to train a model offered into Australian creative production, the provider bears primary responsibility as the party capable of substantiating the lawful basis for that use. The Committee should consider what level of provider transparency or auditable evidence is required to make that responsibility meaningful.
Keep downstream responsibility with the production for material it supplies, the human authorship it claims, performer and likeness consent, and rights in material it incorporates.
A production should remain responsible for material it supplies for prompting, reference, fine-tuning or conditioning; for the human authorship of the work it claims; for performer and likeness consent; and for the rights in material it actually incorporates into the finished production.
Do not collapse the upstream and downstream copyright questions (training legality, output infringement, human authorship); address each where the relevant evidence exists.
Do not collapse the upstream and downstream copyright questions. Whether material was lawfully used to train a model, whether a particular output infringes an earlier work, and whether an AI- assisted output contains sufficient human authorship to attract copyright are different questions. Policy should address each at the point where the relevant evidence exists.
Give authorship a workable test by clarifying the human contribution required for copyright in AI-assisted work, without requiring a new universal registry.
Give authorship a workable test. Australian law offers no settled position on how much human involvement is required for copyright in AI-assisted work [6][7]. Clarity on the relevant human contribution would help writers, producers, publishers and funders preserve the evidence they already generate through drafts, revisions and ordinary production records, without requiring a new universal registry.
Use existing performance and likeness consent and rights machinery with explicit permissions for generated performances; consider whether remedies remain practical for resemblance without deliberate copying at large scale.
Use existing consent and rights machinery for performance and likeness, but make the scope explicit. Synthetic voice, digital doubles and other generated performances should sit within clearly expressed performer consent and contractual permissions [4][11]. The harder question for policy is whether existing remedies remain practical where resemblance can arise without deliberate copying and at very large scale [5].
Use a shared vocabulary for degrees of AI use and avoid regulating unlike uses as though they were the same.
Use a shared vocabulary for degrees of AI use and avoid regulating unlike uses as though they were the same.
Where an assurance requirement is imposed, require evidence that the control works, tested against known positive and negative cases rather than accepted because it is present.
Where an assurance requirement is imposed, require evidence that the control works. My testing found controls that appeared to operate while measuring nothing, and evaluators whose remaining errors became more persuasive as model capability improved. Any required technical control should be tested against known positive and negative cases, not accepted merely because it is present.