'It's in the vibe': vibe coding, copyright ownership and how it changes software contracting in Australia

In a development that would undoubtedly confuse the fictional Dennis Denuto, vibe coding is a hot topic. But what is it and how does it change things?

In short vibe coding means using AI to write software. Increasingly, developers are moving away from manual coding and utilising tools such as GitHub, Copilot, ChatGPT, Claude and Gemini to generate substantial portions of software code from natural language prompts.

Vibe coding minimises the need for a human to write the code, shifting the role of developers to describing objectives, testing and fine tuning outputs and integrating generated solutions.[1]

For many organisations, the benefits are obvious. Vibe coding can dramatically increase development speed, reduce costs and enable subject matter experts - not computer specialists - to create software applications. However, it also raises complex questions regarding copyright and contractual risk allocation.[2]

Australian copyright law has developed on the basis that creative works are produced by humans – we discuss this point further below.

Generative AI changes that. More importantly, where software is generated with little or no human input apart from basic prompting of an AI system, uncertainty arises over whether copyright protects the code, who owns it and whether software vendors should continue to provide the broad intellectual property indemnities that have traditionally appeared in software contracts.

These issues are likely to draw greater attention as AI-assisted development shifts from an experimental practice to a standard development methodology, with research from major coding platform ‘Stack Overflow’ suggesting that 84% of developers globally have used or plan to use AI coding tools in 2025.[3]

What is vibe coding?

The term was coined by renowned computer scientist and former OpenAi co-founder Andrej Karpathy in 2025 (and has no connection to that iconic Australian movie) and refers to a process of software development where, rather than manually writing the code line by line, the developer describes the outcome they want – or the ‘vibe’ – and relies on AI to generate most of, if not all the code. The developer's contribution becomes one of direction, curation and quality assurance rather than direct authorship.[4]

This creates a legal tension because under Australian law, copyright subsists only in works created by a human author. The more coding activity performed by the AI, the more difficult it becomes to identify the human creative contribution that copyright law traditionally protects.[5]

Copyright protection under Australian law

Copyright in Australia is governed by the Copyright Act 1968 (Cth) (the Act).

Section 32 of the Act provides that copyright subsists in original literary works. Computer source code has long been recognised as a literary work for the purposes of the Act.

However, Australian copyright law does not protect ideas; it protects the original expression of those ideas. As a general rule, copyright protection emphasises the need for independent and intellectual human effort in the creation of copyright works.

Relevant key decisions

In IceTV Pty Ltd v Nine Network Australia Pty Ltd (2009) 239 CLR 458 (IceTV), the High Court confirmed that copyright protection depends on independent intellectual effort being expensed by a human author. The mere process of or mechanical expense of time and effort is not enough.[6]

In a similar judgment, the Federal Court in Telstra Corporation Ltd v Phone Directories Company Pty Ltd (2010) 194 FCR 142 (Telstra) held that ‘the sweat of the brow’ does not, of itself, attract copyright protection under the Act. The work must originate from the intellectual effort of a human author.[7]

The significance of Telstra cannot be overstated as it demonstrates that Australian copyright law requires an identifiable human author. Where work emerges from automated processes and no individual can be identified as the creator, copyright protection may fail altogether.[8]

The logic of these decisions presents an obvious challenge for AI-generated software.

Can copyright subsist in AI-generated code?

Australian legislation currently contains no specific provisions dealing with AI-generated copyright works.

As a result, existing principles must be applied. As it stands, an AI system cannot be the author of works. Therefore, without sufficient independent intellectual effort from a human, copyright is unlikely to subsist in works generated by AI.

Different forms of AI use

Scenario 1: Pure AI Generation

Suppose the developer enters a simple prompt requesting a complete software system and the AI generates the application in its entirety without further modification.

The difficulty is identifying the author. The AI is not recognised as a legal person and cannot own copyright. Equally, the user has not contributed sufficient intellectual effort to qualify as an author.[9]

Applying the reasoning in Telstra, there is a credible argument that copyright may not subsist in the resulting software at all.

Scenario 2: AI-Assisted Development

The more common scenario involves extensive human review, modification, testing and integration.

The developer drafts the prompts to instruct the AI, evaluates outputs, rejects undesirable outcomes, rewrites code, tests and then incorporates generated components into a larger architecture.

In this scenario, there is a stronger argument that sufficient human intellectual effort exists to satisfy the requirements identified in IceTV.

Scenario 3: Collaborative Human-AI Creation

The most difficult cases lie between these extremes.

Questions arise such as how much editing is required, whether engineering outputs with further prompts constitutes authorship and whether selecting from alternative outputs is sufficiently creative and of course, evidence or proof of effort.

These matters are yet to be determined by Australian courts.

Ownership issues for employers

Historically, ownership of software was relatively straightforward.

Section 35 of the Act generally provides that works created by employees in the course of employment belong to the employer.

However, vibe coding complicates this analysis.

When AI is used in the scenarios discussed above, the developer technically owns nothing, either due to copyright not subsisting in the work and/or the work belonging to their employer.

The issue carries greater magnitude for the employer who, in situations where AI is used without sufficient independent intellectual effort, may assume it has acquired a proprietary software asset when it has nothing.

This distinction matters during mergers and acquisitions, software licensing transactions, venture capital investments and intellectual property due diligence exercises. Being able to prove ownership of the software is critical for:

1. sellers, in realising the value of those assets, and
2. buyers, in terms of their ability to exploit the assets.[10]

Copyright infringement risk

Ownership is only one side of the problem.

Generative AI systems are trained on vast quantities of existing code and content. Questions remain regarding whether AI outputs may include copyright protected material or substantially similar works.  It is difficult if not impossible to check for this so a user must rely on the supplier of the AI service to give it comfort (and a suitable indemnity). That itself may prove a challenge.

Vendors must always be aware of what they have ‘under the hood’ when it comes to the provision of software. A vendor using software substantially generated by AI may not be able to confidently represent that all components are protected by copyright or are exclusively owned.[11]

It may also lead to disruptive and costly disputes with the real owners of the software or orders forcing the vendor to cease its use which can be catastrophic if it comprises a substantial part of their service offering.

Emerging contractual responses

Disclosures

AI use disclosure clauses can be used to put external users, such as customers or potential buyers, on notice and promote transparency that products/AI solutions have been created using AI.[12]

Making such disclosures may of course trigger enhanced due diligence requirements as buyers (of software or businesses) having been made aware, will need to take the reasonable steps to inform themselves of any matters that will impact the decision to buy.

Warranties

With increased use of AI, consumers will want assurances that a vendor/owner owns or controls the software, has the right to sell or license it and that the use of the software will not infringe third-party rights. Giving these warranties may be riskier.

Governance recommendations

Organisations adopting vibe coding should implement clear AI development policies, require meaningful human review, maintain records demonstrating human contribution and update contractual documentation.

Conclusion

Vibe coding represents a profound shift in software development. While it delivers substantial productivity benefits, it exposes uncertainty regarding authorship, ownership and infringement. The implications extend directly to software transactions and contractual risk allocation. Until legislators or courts provide clearer guidance and governance, care should be taken in using vibe coding techniques and carefully drafted contractual protections will remain the primary mechanisms for managing the legal risks associated with AI-assisted software development in Australia.

Note: The title of this article is taken from The Castle, Directed by William Castle (Village Roadshow, 1997).

[1] IBM, 8 April 2025, https://www.ibm.com/think/topics/vibe-coding.

[2] Ibid.

[3] The Conversation, 30 July 2026, https://theconversation.com/vibe-coding-is-fun-and-easy-but-theres-a-major-catch-287693.

[4] Ibid.

[5] University of Melbourne, 2026, https://copyright.unimelb.edu.au/shared/using-copyright-material/ai-and-copyright.

[6] IceTV Pty Ltd v Nine Network Australia Pty Ltd (2009) 239 CLR 458

[7] Telstra Corporation Ltd v Phone Directories Company Pty Ltd (2010) 194 FCR 142

[8] Ibid.

[9] Viva Tech, 3 December 2026, https://vivatech.com/news/who-owns-ai-generated-content-the-answer-is-complicated.

[10] The Code Registry, 23 July 2026, https://thecoderegistry.com/how-is-the-increase-in-ai-assets-going-to-impact-intellectual-property-due-diligence-in-ma-transactions/.

[11] Ibid.

[12] OAIC, 17 January 2025, https://www.oaic.gov.au/privacy/privacy-guidance-for-organisations-and-government-agencies/guidance-on-privacy-and-the-use-of-commercially-available-ai-products.