Generative AI is trained on vast amounts of information (in the case of ChatGPT it is text material, image-based GenAI may use images, etc.) This information is often Public Domain, or otherwise legally-licensed material, however, a lot of the information is from Copyrighted sources. Ask ChatGPT to summarise Harry Potter books, and it would not be able to give a breakdown of the book's plot, unless it had access to the Copyrighted Work in its training data.
See the current cases of authors v OpenAI, "Two authors are suing OpenAI for training ChatGPT with their books. Could they win?" (, 2023).
Note: This is quite complex in terms of Copyright law, given the models do not actually contain the Works themselves, and have only been trained on the patterns found within the Works (with the Works then deleted from the model.) This means the GenAI may be likely to generate an acurate summary of Harry Potter but this is due to a set of learned, probabilistic relationships between words, and not a direct copying of the Work. It is also just as likely the GenAI will "hallucinate" a likely, but incorrect summary of the Work.
Some of this Copyrighted material may have been licensed for this use, and there are arguments this training use may come under Fair Dealing provisions.
Can works generated by Gen AI have Copyright protection? And who owns the copyright in Works created using Gen AI?
Obviously, the Australian Copyright Act 1968 doesn’t explicitly mention Gen AI, however, the existing provisions of the Act can give us some guidance around what Copyright protects, and under what circumstances it applies.
To be a protected Work (or Subject-Matter Other than Work), all literary, dramatic, musical and artistic works must:
Copyright protection in Australia is free and automatic as soon as original content (of “independent intellectual effort”) is fixed in a material form (such as when a song is recorded, an artwork is painted, or you take a picture.) It does not need to be registered, applied for or paid for.
However, regardless of this expression, as the law currently stands, if a Work is generated by AI and does not possess sufficient human input in its creation, then it is not granted protection under Copyright.
This means that such Works (and Subject-Matters Other than Works) do not have Copyright protections, including from:
However, international cases may yet provide further insight toward the future of Copyright protection for Works partially generated by AI. This is due to humans co-creating a Work with Gen AI being considered as enough of an “intellectual effort” to qualify for protection. At this time, it is unclear what extent of co-creation is required for coverage under Australian Copyright law.
The current advice from ArtsLaw is that “when a Work is generated with the help of AI, human authors who contributed “independent intellectual effort” to the creation of the Work should be credited as authors.” Whilst the Copyright protection is not yet settled law, this helps identify creators for attribution, and future Copyrights purposes.
ECU Library further suggests that when co-creating with Gen AI to keep records of all creative in-put, prompts, Gen AI outputs, and editing as evidence of human “intellectual effort” for future Copyright claims.