No Register, Plenty of Copyright Claims

2026-11-02

An Australian companion to Copyright Claims Are Cheap. Challenging Them Isn't., which examines Cory Doctorow's prediction that generative AI will bring back copyright formalities in the United States.

Australia presents an interesting test of that argument. It has no copyright registration system at all. It also has established case law denying protection to works produced by highly automated processes, statutory presumptions that help copyright claimants, and a publishing industry that is already introducing AI disclosures.

Those facts point in different directions. But they don't suggest that corporate claims to intellectual property are about to evaporate.

No register, no need to prove ownership in advance

The Australian Attorney-General's Department is explicit: copyright protection is automatic, with no registration system or particular document required to establish oneself as creator and owner. A copyright notice isn't compulsory either.

For an original literary, dramatic, musical or artistic work, copyright generally depends on human authorship involving independent intellectual effort. Substantial AI assistance doesn't automatically rule it out; how much human creative contribution is sufficient remains an unsettled question for modern generative systems. The Arts Law Centre of Australia explains the distinction.

A publisher can therefore put a copyright notice on an AI-assisted book and sell it without first obtaining approval from a copyright registrar. That notice cannot create rights that the law doesn't recognise. But the marketplace has no routine mechanism that tests the claim before the work goes on sale.

This makes a sudden, universal requirement to certify the human origin of every word or pixel seem unlikely.

Australian courts can be tough on automated works

The strongest qualification comes from two cases that predate generative AI.

In Telstra Corporation Ltd v Phone Directories Co Pty Ltd (2010), the Full Federal Court held that copyright did not subsist in Telstra's White Pages and Yellow Pages directories. Their compilation relied heavily on automated processes, and Telstra could not establish the required original human authorship of the directories as claimed.

In Acohs Pty Ltd v Ucorp Pty Ltd (2012), the Full Federal Court rejected claims to copyright in particular HTML source-code outputs assembled automatically by a computer system. Human beings had written software and entered data, but that didn't make them authors of each generated output.

These weren't LLM cases. They also don't mean every computer-assisted work is unprotected. They show something narrower and important: once an Australian court seriously examines a disputed claim, extensive automation and uncertain human authorship can be fatal.

An AI-generated work isn't made copyrightable by calling it proprietary.

But what happens when the facts are disputed?

The Copyright Act 1968, sections 126–128, contains evidentiary presumptions that complicate the picture.

Section 126 presumes that copyright exists if the defendant doesn't dispute its existence. That presumption falls away when subsistence itself is put in issue. So an allegation of inadequate human authorship can matter.

But other provisions may still assist a claimant. Section 127 presumes, unless the contrary is established, that a person named as the author on a published work really is its author. Section 126B supplies rebuttable presumptions concerning ownership statements on copies and documented chains of ownership. Section 128 can, in specified circumstances, presume both copyright subsistence and publisher ownership.

These rules don't magically confer copyright on machine-generated expression. Nor is there a general rule that anyone alleging AI use must prove it beyond doubt: the plaintiff may ultimately fail if it cannot establish a valid right.

Rather, the result depends on the particular challenge, the evidence, and which presumptions apply.

Imagine a publisher saying a named writer created a book, while a competitor alleges the text was generated mostly by an LLM. Without convincing evidence on either side, the outcome isn't automatically 'public domain'. Equally, a publisher unable to establish relevant human authorship cannot assume it will win merely because it printed © on the cover.

The commercial implication is familiar: challenging the claim may be expensive and risky even when there is a respectable argument that the claimed copyright doesn't exist.

The new paperwork is already here

There is genuine Australian evidence for part of Doctorow's prediction.

In September 2026, Melbourne University Publishing adopted a written AI policy requiring authors to disclose generative-AI use in producing their work. It distinguishes ordinary spellchecking and grammar assistance from substantive generative use and places responsibilities on authors concerning manuscripts and data uploaded to AI systems.

An ABC investigation published on 28 October 2026 reported publishers updating author contracts, scrutinising submissions and exploring human-authorship certification. Separately, the Australian Society of Authors released guidance on AI clauses in June 2026 to help writers and illustrators negotiate terms concerning AI licensing.

Meanwhile, the federal government's Copyright and Artificial Intelligence Reference Group is considering licensing, uncertainty around AI-generated outputs, and cheaper enforcement mechanisms. Its consultations are not themselves new statutory formalities.

These examples demonstrate growing private administration. But they are targeted: a publisher wants trustworthy manuscripts; an author wants control over AI licensing; a buyer wants confidence in an asset.

They do not establish a general Australian requirement for everyone publishing a work to document the contribution of every person and model.

And the information collected can remain entirely private.

What will the market believe?

Ultimately, the question is the same as in the US companion essay: will AI authorship uncertainty materially alter what buyers, licensees, investors and competitors believe a business owns?

For some assets, yes. A company selling purportedly exclusive rights over largely autonomous AI outputs may have serious trouble defending that exclusivity. Australian case law gives challengers reason to investigate.

For a publisher with a mixed catalogue of human-authored and AI-assisted books, contracts, brands and established customers, the position is different. Its business doesn't suddenly become worthless because some new work contains AI-generated material. The value of its existing rights, and the cost of replacing its products, matter as much as the theoretical scope of copyright in one disputed output.

So Australia may see more contracts, warranties, disclosures and selective evidence-keeping. In publishing, it already is.

That is meaningful, but it isn't a general return to public copyright formalities. It is the ordinary market adapting its claims to a new source of legal uncertainty.

Australia's lack of a copyright register makes claims easy to make. Its courts can make some of those claims difficult to enforce. Between those two facts lies a great deal of commercially valuable ambiguity.

AI disclosure: Generative AI was used in producing this article. How much? I'm not telling. ;-)