Submission 547
Alt.vfx
Submitter group: Companies.
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 AI disclosure needs verifiable process evidence alongside watermarking and provenance. Recommends audit-ready copyright declarations, domestic technical evaluation and graded labels with privacy safeguards. States a commercial interest in verified-disclosure legislation and research, while saying its technology is unfinished and partly vulnerable to human-in-the-loop laundering. Says its recommendations do not require engaging any particular provider.
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 any copyright licence, opt-out or consent mechanism will rest on claims made by the parties themselves about what they did.
Whichever copyright mechanism the committee favours (a statutory licence, an opt-out, a consent register or a remuneration scheme), it will rest on claims made by the parties themselves about what they did.
Argues that cryptographic provenance shows a record was signed and unaltered, but not that its content is true.
Cryptographic provenance proves that a record was signed by a particular party and has not been altered since. It does not prove that what the record says is true.
Argues that content-based detection of AI-generated media, without process evidence, is not yet fit for purpose.
Content-based detection of AI-generated media, absent the additional context of process evidence, is not yet fit for purpose.
Says, to its knowledge, no Australian technical evaluation has been commissioned to underpin forthcoming AI standards.
The Prime Minister announced Australian Standards for AI on 15 July 2026, alongside a new Office of AI within the Department of the Prime Minister and Cabinet, with legislation expected in early 2027 [17]. However, to our knowledge no Australian technical evaluation has yet been commissioned to underpin it:
States its position that watermarking or signing shows whether content was generated, process evidence shows how much was human, and a realistic framework needs both.
Our position is simply that watermarking or cryptographic signing answers whether content was generated, but process evidence answers how much of the work was human, and a realistic framework needs both, because neither one substitutes for the other.
States that its technology is unfinished and offers evidence that a verification layer is buildable and testable.
We are not claiming to offer the committee a finished and directly applicable technology as of today: we are simply offering evidence that the verification layer is buildable and testable,
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Design copyright mechanisms (licences, opt-outs, consent registers or remuneration schemes) around presently unverifiable usage and human-authorship declarations; require forms auditable against process evidence as verification methods mature.
Any copyright mechanism the committee recommends (licence, opt-out, consent register or remuneration scheme) should be designed on the explicit assumption that the declarations made under it — of what was used, and of how much of a work is human — cannot presently be verified, and should require those declarations to be made in a form that can later be audited against process evidence, as verification methods mature.
Have a framework distinguish declared from verified disclosure and leave room for independent verification against process evidence, with a foundation on which a 'verified' finding can be challenged or revoked.
A framework should distinguish in its own terms between declared and verified disclosure, and should leave room for independent verification against process evidence — inclusive of a technical foundation on which a “verified” finding can be challenged or revoked.
Do not mandate AI detection as the basis of any obligation.
AI detection should not be mandated as the basis of any obligation, for the reasons set out in section 3.
State explicitly that absence of provenance data is not evidence of inauthenticity.
It should be stated explicitly that absence of provenance data is not evidence of inauthenticity, so that this does not become the default assumption.
Commission a domestic, modality-specific technical evidence base on watermarking, provenance and detection before standards and obligations are settled, and support domestic research collaboration between institutions and industry.
Australia should commission a domestic, modality-specific technical evidence base — evaluating the state of the art in watermarking, provenance and detection — before the relevant standards and obligations are settled rather than after, and should support the domestic research, including collaboration between research institutions and industry, needed to build and retain that capability in Australia.
Attach disclosure obligations to process and method rather than creator identity, permitting minimum-disclosure and pseudonymous credentials to preserve privacy.
Disclosure obligations should attach to process and method rather than to creator identity, and should permit minimum-disclosure and pseudonymous credentials which allow for privacy to be maintained when needed.
Avoid a universal chain of custody built into every creative tool by default.
A perfected provenance chain is also a perfected tracking chain, and the C2PA specification team maintains its own harms model recognising that risk [23]. For the same reason, we would not support a universal chain of custody being built by default into every creative tool.
Make labelling graded and per-component rather than binary, aligned with the EU's fully AI-generated versus partially AI-modified distinction, building on existing international standards rather than an Australia-specific mechanism.
Labelling should be graded and per-component rather than binary, aligned with the European Union’s distinction between fully AI-generated and partially AI-modified material [24], and Australian settings should build on existing international standards rather than create an Australia-specific mechanism
Consider the AI Safety Institute as a natural home for an Australian technical evaluation of watermarking, provenance and detection techniques.
there appears to be no provision for an Australian assessment of whether candidate watermarking, provenance or detection techniques are fit for purpose, in any modality. The AI Safety Institute would seem a natural home for such work.
In the attached Senate submission, asks for specific expertise and dedicated consultation on First Nations cultural knowledge.
Our submission also does not address the position of First Nations cultural knowledge, much of which sits outside copyright and is at risk of falling outside a transparency framework’s protections altogether. We raise it only to say that it requires specific expertise and dedicated consultation,
In the attached Senate submission, proposes commissioning necessary modality-specific standards extensions now, while standards remain open to Australian research contributions.
Our recommendation #6 follows from this directly: the evidence base and necessary extensions to the relevant standards need to be commissioned now, while the standards themselves remain open to Australia’s research contributions, and these should be modality- specific, because what holds for still images does not hold for audio, video or text.
In the attached Senate submission, asks that outputs of process-evidence validation carry an appropriate label denoting a technically verified claim, with a foundation for challenge or revocation.
our recommendation #2 asks only that the framework leave room for such validation technology based on process evidence to be developed, and for the output of such a validation process to carry an appropriate label denoting that the claim is technically “verified” (inclusive of requiring a technical foundation for such validation to be challenged or revoked).
In the attached Senate submission, argues legislation should combine provenance, education, policy and detection, rather than treat provenance alone as sufficient.
We would also add, following the position taken jointly by the United States’ National Security Agency, the Australian Signals Directorate’s Australian Cyber Security Centre, the Canadian Centre for Cyber Security and the United Kingdom’s National Cyber Security Centre [41], that provenance, education, policy and detection are needed together: provenance alone is not sufficient, and should not be legislated as though it were.