Australia and
China approach those questions through different legal and institutional
systems. That difference should not be minimised. It does not, however,
eliminate every opportunity for constructive cooperation. A workable bilateral
agenda need not begin with the harmonisation of national laws or agreement on
every contested principle. It can begin at the practice layer: promoting shared
literacy, maintaining clear consent records, establishing reliable provenance,
conducting proportionate risk assessment, embedding human review and implementing
governance tools that small organisations can actually use.
The central
proposition of this article is that responsible AI cooperation should extend
beyond laboratories, major firms, and high-level forums. It should reach the
grassroots institutions where creative work is produced, taught, licensed,
distributed, and preserved. The ACCESS framework – Access, Comprehension,
Consent, Evaluation, Safeguards, and Stewardship – is an original framework
proposed by the author of this article. It is not a statutory standard nor an
adaptation of any single external framework. Rather, it synthesises recurring
responsible AI principles found in Australian government’s guidance, Chinese
generative AI regulations, UNESCO and OECD principles, and the practical rights-related
questions that arise in the music industry and other creative sectors. Its
purpose is practical implementation: to turn broad principles into a sequence
that a small creative organisation can understand and repeat.
Different systems, useful points of convergence
Australia’s
current approach combines existing technology-neutral laws with practical
governance guidance. The Australian Government’s Guidance
for AI Adoption outlines six essential practices for safe and responsible
AI governance, while the National AI Plan emphasises broad
adoption, capability-building, responsible practice, and the continued
application of existing laws.
China has
adopted more directive rules for public-facing generative AI services. The Interim Measures for the Administration of Generative AI
Services address lawful training data, intellectual property,
personal information, service security, and provider responsibilities. The Measures for Labelling AI-Generated and Synthetic Content
introduce explicit and implicit labelling requirements for AI-generated text,
images, audio, video, and virtual scenes, which have been in force since
September 2025.
These are not
equivalent regimes. Australia’s model remains more distributed across privacy,
consumer, copyright, employment, safety, and other laws, while China’s
framework places more explicit obligations on defined providers and
content-distribution processes. Yet both systems recognise that AI adoption
requires accountability, risk controls, transparency, and attention to the
interests of affected individuals. The same themes appear in UNESCO’s Recommendation
on the Ethics of Artificial Intelligence and the OECD AI
Principles, which provide a useful neutral vocabulary for
human-centred, transparent, robust, and accountable AI.
The contrast also creates a reason for cooperation in its
own right: comparative learning. Australia and China can examine how different
governance tools perform against similar creative-industry risks – rather than
treating regulatory differences solely as obstacles. Australia can study the
practical effects of more prescriptive mechanisms such as synthetic-content
labelling and defined platform responsibilities, while Chinese stakeholders can
examine Australia's risk-based governance, organisational accountability, human
oversight, and implementation guidance. The objective would not be to replicate
one system in the other, but to identify which safeguards are transferable and
which approaches most effectively improve provenance, reduce misuse of voice
and likeness, address copyright and attribution concerns, and preserve opportunities
for responsible creative innovation. In this sense, bilateral cooperation can
function as a practical testing ground for learning what works across different
regulatory environments.
Why the creative sector needs its own implementation layer
Creative
industries are not merely another field of AI deployment. They trade in
identity, expression, cultural memory, reputation, and public trust. A singer’s
voice can be both a marker of personal identity and a commercial asset. A
designer’s body of work can be both training material and a source of livelihood.
A community arts organisation may handle recordings, photographs, traditional
knowledge, and information about minors without the benefit of a dedicated
legal, privacy, or cybersecurity team.
Australia’s
copyright framework protects works, sound recordings, films, broadcasts, and
moral rights, while policy questions surrounding AI inputs, transparency,
generated outputs, licensing, and enforcement remain the subject of ongoing
debate and development. The Copyright and Artificial Intelligence Reference Group
is a standing mechanism for engagement on copyright-AI issues, including
participation by stakeholders from the music, screen, voice-actor, publishing,
technology, and cultural sectors.
China’s
judicial practice illustrates why identity rights must sit beside copyright
analysis. In a case highlighted by the Supreme People’s Court, a voice-over artist
discovered that recordings of her voice had been transferred and used to
generate an AI voice without her consent. The court found that an infringement
of her voice rights had occurred, because the synthetic voice remained recognisably
connected to her vocal characteristics. For musicians and performers, the
lesson is direct: authorisation to make or distribute a recording does not
automatically confer permission for a voice to be cloned, modelled, or used to
generate new performances.
A bilateral
creative-AI agenda must therefore avoid collapsing every problem into a single
label such as “copyright.” Rights may arise from copyright, performers’ rights,
contracts, privacy, consumer protection, personality or voice rights, platform
rules, and sector-specific content regulation. The practical task is to help
creators identify which questions must be answered before material crosses a
platform, organisation, or national border.
The ACCESS framework
To translate broad principles into practical action, this
article proposes the ACCESS framework: Access, Comprehension, Consent,
Evaluation, Safeguards, and Stewardship. Rather than prescribing legal rules,
it provides a practical structure for responsible AI adoption in the creative
sector.
Access
Responsible
adoption begins with meaningful access to tools, training, and support. Access
is not achieved merely because a free AI service is available online. Creators
also need affordable connectivity, accessible interfaces, appropriate language
support, and guidance suited to small organisations. Australia’s AI
Adopt program, offers a practical model by supporting small and medium
businesses with applied AI assistance. China’s expanding AI-literacy
initiatives show a parallel commitment in widening capability. A
creative-industry pilot could adapt these approaches for music schools,
cultural associations, independent labels, design studios, screen producers,
and community media organisations.
Comprehension
People should
understand how an AI system works, what information it receives, how its
outputs are generated, and where uncertainty remains. Creative AI literacy
should cover prompting and productivity, but also hallucinations,
synthetic-media detection, bias, licensing, privacy settings, disclosure, and
the limits of automated decision-making. Training should be available in plain
language and tailored to a broad range of participants, including those without
technical expertise. Participants should also recognise that AI-generated
outputs may contain errors, omissions, or fabricated information and should be
able to distinguish verified facts from model-generated suggestions.
Comprehension
is particularly important in cross-border projects because a workflow that
appears routine to one jurisdiction may carry different disclosure, data, or
content obligations in another.
Consent
Consent must be
specific enough to be meaningful. Permission to record a performance does not
automatically extend to train a model, clone a voice, generate new
performances, or transfer files to another provider. A model clause or consent
record should clearly identify the material, purpose, system, duration,
territorial scope, recipients, compensation arrangements, revocation process,
and whether synthetic derivatives are permitted.
Where personal
information is involved, organisations should also ensure compliance with
privacy law. Australia’s privacy guidance on generative-AI training emphasises
that broad privacy-policy consent may not be sufficient for complex AI training
uses and that people need meaningful information about how their data will be collected,
used, and disclosed.
Evaluation
Evaluation asks
whether an AI use is appropriate before it becomes routine. Small organisations
do not need a laboratory-scale assurance program, but they do need a simple
impact assessment. It should identify the purpose, users, data, rights holders,
foreseeable harms, content-regulation risks, the responsible human reviewer,
vendor dependencies, and clear criteria for halting or modifying the activity
if risks become unacceptable.
Higher-risk
uses – such as cloning a performer, generating political or health-related
content, making decisions about employment, or processing children’s data – require
more rigorous assessment and oversight or it may be deemed unsuitable for a
community pilot.
Safeguards
Safeguards
convert policy into practical controls and accountability measures. Examples
include approved-tool lists, restricted data categories, rights-clearance
checks, visible AI disclosures, provenance metadata, secure storage, human
approval before publication, a complaints channel, and an incident log.
China’s
labelling framework offers a concrete example of end-to-end governance
responsibilities spanning content generation, labelling, and distribution.
Australia’s guidance offers a complementary risk-management process. A joint
toolkit should document both, clearly marking which controls represent good
practice and which arise from mandatory legal or regulatory obligations.
Stewardship
Stewardship
continues after a tool is adopted. Someone must own the policy, update the risk
assessment, maintain authorisation records, respond to complaints, and preserve
evidence when misuse occurs. Stewardship also means resisting unnecessary data
collection and retaining human authority over culturally or personally
significant decisions.
A small organisation
should be able to answer five questions at any time: What system are we using?
For what purpose? What material is being uploaded and under what authority? Who
reviewed the output? How can an affected person challenge, correct, or seek
review of a decision or outcome?
Existing cooperation shows the problem is real
Recent
developments have made the Australia-China creative AI connection more tangible.
In April 2026, the Australian AI Music Alliance announced a strategic
partnership with Tomato Music, a Chinese audio platform under ByteDance, to
launch the Tomato
AI Music Hit Awards for original AI music creators in Mainland China. The
initiative links an Australian AI music industry organisation with a Chinese
distribution platform to support AI-assisted creation, discovery, evaluation,
and circulation. It demonstrates that bilateral collaboration is already moving
beyond general cultural exchange into AI-enabled creative production and
distribution.
At the same time, China's creative industries are testing AI
within real-world production workflows. The 2026 Shanghai International Film
Festival launched its AI
Backlot program, pairing filmmakers with AI creators to complete end-to-end
AI-image production processes while documenting prompts, workflows, process
materials, and lessons for industry learning. Although this initiative is not
itself an Australia-China bilateral program, it provides a contemporary example
of the production environment in which future Australian-Chinese screen
collaborations may operate. Together, these developments sharpen the governance
questions identified in this article: how consent, provenance, attribution,
disclosure, human review, and rights clearance should travel with creative
material across platforms and borders.
The
institutional pathway for practical cooperation also already exists. The National
Foundation for Australia-China Relations’ 2026–27
grant program identifies dialogue, practical cooperation, industry
engagement, and arts-and-culture exchange as intended outcomes. A creative-AI
literacy or governance pilot would therefore not require the creation of an
entirely new diplomatic channel; it could build on existing people-to-people
links, institutional partnerships and sectoral cooperation.
There is also
evidence that the creative AI challenge is no longer theoretical in Australia
itself. Creative Australia notes a significant
increase in the use of generative AI systems for content creation and has
issued principles designed to ensure that human creativity remains central.
Combined with China’s more prescriptive requirements for public-facing
generative AI and synthetic-content labelling, this highlights the practical value
of bilateral AI literacy and governance initiatives: collaborators may use the
same tools but face different expectations about authorisation, disclosure,
provenance, and distribution across jurisdictions.
Three initiatives that can start small
First,
establish an Australia-China Creative AI Literacy Exchange. A twelve-month
pilot could pair educators and cultural organisations to deliver short
bilingual modules on responsible prompting, copyright and related rights, voice
and likeness, privacy, synthetic-content labelling, and human review.
Participation should include independent creators and regional or community
organisations, not only universities and technology companies.
Second, develop
a Creator Consent, Provenance, and Rights Playbook. The playbook should include
plain-language intake questions, a consent-record template, a rights-clearance
checklist, a synthetic-content disclosure template, a vendor due-diligence
form, and a rapid-response procedure for unauthorised synthetic replicas or
imitations. Each item should contain jurisdiction specific notes explaining
where Australian and Chinese rules diverge. It should not present itself as a
substitute for legal advice.
Third, create a
Small Creative Organisation Governance Kit based on ACCESS. The kit should let
an organisation inventory its AI uses, assign an accountable owner, classify
risks, approve or prohibit use cases, document incidents, and review the policy
every six months. A pilot should test the kit with a small group of music,
media, screen, design, and community-education organisations and publish an
anonymised lessons-learnt report.
A twelve-month operating model
The first three
months should be devoted to co-design. A small steering group should include
independent creators, educators, cultural organisations, rights and privacy
specialists, and people familiar with each country’s regulatory environment.
The group would select low-risk pilot uses, agree on bilingual terminology,
identify materials that may lawfully be used, and define clear boundaries and
prohibited practices. Higher-risk activities, such as training a
general-purpose model on participant works, cloning an identifiable person
without specific authorisation, and processing sensitive personal information
in public AI tools should be excluded from the first phase.
Months four
through eight should test the literacy modules and governance kit through
supervised workshops. Participants could use AI for tasks such as translation
of approved promotional copy, metadata drafting, accessibility support,
brainstorming, subtitle preparation, and audience research using non-sensitive
information. Each use would generate a short record covering the purpose,
source material, rights basis, AI system used, human reviewer, disclosure
decision, and outcome. The record is not bureaucracy for its own sake; it
creates an auditable trail that can be reviewed when a mistake, complaint, or
rights question arises.
Months nine
through twelve should focus on evaluation and revision. Participants should
report which controls were understandable, which created unnecessary burden,
which risks were missed, and whether the process changed their willingness to
adopt AI. The steering group should publish an anonymised report, update the
templates, and recommend whether a larger exchange is justified.
Any expansion
should remain modular: a common ACCESS core, jurisdiction-specific legal notes,
and sector modules for music, visual arts, screen, publishing, design, and
cultural heritage. This staged approach also aligns with the international
cooperation and capacity-building direction reflected in China’s 2026 AI Cooperation and Development Action Plan,
while keeping the project bounded, evidence-based, and accountable to
participating creative communities.
Constraints that cooperation must face honestly
Regulatory
differences are the first constraint. A shared toolkit cannot promise that
compliance in one country establishes compliance in the other. It should therefore
combine a common core of responsible practices with separate jurisdiction
modules and clear referral pathways to qualified local advice.
Content
regulation is the second constraint. China’s requirements for public-facing
generative AI and synthetic-content labelling are more prescriptive than
Australia’s current principles-based and technology-neutral approach. Projects
must define which content will be created, where it will be distributed, which
platform is responsible for labelling and disclosure, and which local rules
apply. Cultural exchange should not become a route around domestic content
obligations.
Intellectual
property and data transfer create a third constraint. Rights status may differ
across recordings, compositions, images, performances, archives, and
traditional cultural expressions. Cross-border movement of personal information
or sensitive creative archives may trigger additional obligations. Pilot
projects should minimise data collection and sharing, use licensed or
participant-supplied materials, avoid training general-purpose models on
participant content, and document data retention, deletion and consent-withdrawal
procedures.
Finally,
cooperation depends on trust. Translation errors, unequal bargaining power,
inaccessible contracts, and uncertainty about downstream model use can
undermine participation. Governance should therefore include bilingual notices,
independent creator representation, published decision criteria, a complaints
process, and transparent reporting of both outcomes and incidents.
A measurable pilot, not another declaration
Success should
be measured by practice: the number and diversity of creators trained;
improvements in participants’ ability to identify and manage risks; the
proportion of pilot activities supported by documented authority and human
review; response time for complaints; number of unauthorised uploads prevented;
and whether organisations continue using the toolkit after the pilot. The aim
is not to prove that Australia and China have the same AI system. They do not.
The aim is to test whether shared human-centred practices can protect creators
and improve responsible adoption across different systems.
Guardrails
matter, but they become meaningful only when people can understand and use
them. By placing access, comprehension, consent, evaluation, safeguards, and
stewardship in the hands of creators and small institutions, Australia-China
cooperation could move from abstract agreement to accountable practice. The
creative economy is an appropriate place to begin because cross-border
collaboration already exists, AI use is expanding, the risks are visible, and
success depends on preserving the human creativity, identity and agency that
technology is meant to serve.
