Who Owns the Copyright to AI-Generated Content Your Business Creates?

Detailed Explanation

Australia's Copyright Act 1968 requires a human author — a work has to reflect human creative choices to qualify for protection, and Australian courts have already tested this directly against machine-produced material. In Telstra Corporation Ltd v Phone Directories Company Pty Ltd (2010), the Federal Court considered whether telephone directories compiled largely through an automated process were original literary works, and found they weren't protected by copyright because no identifiable human author (or authors) could be pointed to for the relevant compilation and expression — the process was too automated for the law's authorship requirement. That reasoning creates a specific, practical gap for AI-generated content: a piece of marketing copy, a product description, or an image generated by an AI tool with no meaningful human input has, on the same logic, no copyright owner at all. It isn't that the business doesn't own it and someone else does — nobody owns it, because purely AI-generated output doesn't meet the authorship threshold Australian copyright protection requires.

This matters most for anything a business wants exclusive rights over. If a competitor copies a blog post that was published as-is from an AI tool with no human editing, there's typically nothing to enforce against them, because there was no copyrightable work to infringe. The same isn't true, though, once a person makes genuine creative decisions on top of the AI draft — selecting which of several AI-generated options to use, substantially rewriting the structure and wording, combining AI output with original human-created material. Applying the same authorship principle from Telstra v Phone Directories, the resulting work is copyrightable to the extent of that human contribution: the AI-generated parts still aren't protected on their own, but the human-authored expression layered on top is.

There's a second, separate risk worth distinguishing from ownership: an AI model's output can occasionally resemble copyrighted material it was trained on closely enough to raise an infringement concern in the other direction — the business publishing AI-generated content could be accused of reproducing someone else's protected work under the Copyright Act, rather than failing to protect its own. This is a different legal question from who owns the output, and it's generally a bigger concern for content published at high volume or used commercially (advertising, branded assets) than for routine internal drafting.

What This Means in Practice

Routine internal and low-stakes content — first drafts, internal documents, day-to-day marketing copy that a person reviews and edits before it goes out — carries limited practical risk either way. Most businesses aren't relying on exclusive copyright over this material, and the human review and editing step that should happen anyway (see how do you use Claude for business tasks and how do you stop AI assistants from making things up) also happens to be what strengthens any copyright claim over the result.

Content the business needs to own exclusively — a branded asset, something being licensed to a partner or client, anything where a competitor copying it verbatim would be a real problem — needs a documented, substantive human contribution: real editorial judgment, not a pass-through of the AI's first draft. Keeping a record of what a person changed, added, or selected supports the claim later if it's ever challenged.

High-volume or commercially significant published content — advertising campaigns, large content libraries, anything produced at a scale where infringement risk compounds — benefits from an occasional review step checking that generated content isn't unintentionally close to existing copyrighted material, on top of the ownership consideration.

Things to Consider

• This is an evolving area of law, not a settled one. Australian courts haven't yet directly ruled on AI-generated content specifically, so Telstra v Phone Directories is the closest existing domestic precedent rather than a case decided on these exact facts — guidance is likely to be refined as AI-specific cases reach Australian courts, so treat the current position as the best available reading today, not a permanently fixed rule, and re-check for anything with significant business value riding on it.
• Vendor terms of service address a different question than copyright law does. Most major AI tools' terms grant the business using them broad rights to use the output commercially, regardless of the underlying copyright status — check the specific vendor's current terms for what they actually promise, since this varies by provider and plan.
• Other jurisdictions take a different approach, which matters if you operate or license content beyond Australia. The United States, the United Kingdom, and the EU apply broadly similar human-authorship requirements, but the specifics and the guidance issued by their respective copyright offices differ from Australia's — a business licensing content into another jurisdiction should verify the position that actually applies there rather than assuming the Australian approach travels automatically.
• This is a specialized legal question for anything of real commercial significance. General awareness of the human-authorship requirement is useful for everyday judgment calls; anything where real money or exclusivity is at stake — a licensing deal, a dispute, a branded flagship asset — warrants a conversation with an IP attorney rather than relying on general guidance.

Common Mistakes

• Assuming AI-generated content is automatically owned by whoever generated it. The more common reality is closer to the opposite — purely AI-generated content with no human creative contribution typically isn't owned by anyone in the copyright sense.
• Publishing an unedited AI draft as a flagship branded asset. The content a business most wants to protect exclusively is exactly the content that most needs substantive human authorship layered on top, not the content most likely to be published as-is for speed.
• Treating a vendor's terms of service as a substitute for understanding copyright status. A vendor's usage rights and the underlying copyright status of the output are two different things — a business can have full rights to use content commercially under a vendor's terms while that same content still isn't independently copyrightable.
• Not keeping any record of human editorial contribution on content that matters. Without a record of what a person actually changed or added, a copyright claim over an AI-assisted work is harder to support if it's ever challenged.