AI Policy & Copyright · September 22, 2026 · 2 min read
Anthropic and OpenAI Ask Australia to Revisit Copyright Rules for AI Training
Both companies have told an Australian review that broad copyright exceptions for AI training lack support, while seeking other ways to access data. The debate turns on licensing, creator rights, and local policy.
Anthropic and OpenAI Ask Australia to Revisit Copyright Rules for AI Training
September 22, 2026
Anthropic and OpenAI have asked the Australian government to reconsider aspects of its approach to copyright and AI training, Reuters reports. Anthropic said it accepts that broad exceptions for AI training have not gained widespread support and have been ruled out by the government. The submissions arrive as countries search for a balance between access to training material and the rights of writers, artists, publishers, and other creators.
A company submission is a policy argument, not a change in law. The current legal framework and any government response must be tracked separately from what AI developers would prefer. The key practical question is how model developers can obtain data lawfully and at scale while creators receive meaningful control or compensation.
The choices policymakers face
One option is to require licenses for particular uses. Another is to create a limited exception with safeguards, or a system that lets rights holders opt out. Each approach has trade-offs. Licensing can compensate creators but may be difficult to negotiate across large datasets. Exceptions can lower friction for research and development but may weaken bargaining power if they are too broad.
Any system also needs to define what “training” covers. A model can process material during training, retrieval, fine-tuning, or user-facing generation. Different uses may create different legal and economic questions. Transparency around datasets and rights-management information can help creators understand what happened to their work.
Why this matters beyond Australia
AI training disputes cross borders. A company may collect material in one jurisdiction, train models elsewhere, and offer services globally. National rules can therefore influence how products are developed and where data partnerships are formed. Businesses using AI should understand the vendor’s approach to copyrighted inputs and what protections apply to generated outputs.
Creators also need practical routes to participate. If licensing mechanisms are too complex or only large rights holders can negotiate, independent artists may receive little benefit. Collective licensing, standard contracts, attribution systems, and clear reporting could help, but require consultation with creator groups.
What to watch
The Australian government’s response, the full submissions, and any proposed statutory language will determine whether this debate changes operational rules. Policymakers should publish impact assessments and hear from creators, technology companies, educators, libraries, and researchers.
The two companies’ submissions show that even leading AI developers recognize political resistance to a sweeping exception. The next phase is about workable alternatives: lawful access, transparent rights, and a process that does not make creators invisible in the AI economy.
Sources
Governance & Safety
Policy, evaluation, security and the control frameworks that make AI deployments defensible to auditors, customers and regulators.
Browse Governance & Safety