Key takeouts

It is predicted that by 2027, 50% of software developers will use AI to augment their coding processes.

The integration of AI in coding raises complex copyright issues, particularly around establishing human authorship and originality.

Businesses should include contractual obligations ensuring human skill and labor in software development agreements and maintain detailed records of human contributions.

In 3 years, 50% of software developers will code using artificial intelligence (AI)¹. With every output being informed by prior entries as AI picks up the grunt work for developers, who really owns the intellectual property?

The software development landscape has seen major changes, as access to generative AI has transformed the industry, allowing developers to use AI to speed up the software development lifecycle. This practice is often referred to as AI-augmented software development.

This article explores copyright law challenges to navigate in AI-augmented software development, including how traditional conceptions of authorship and creativity might apply in the context of AI-generated outputs; and who, if anyone, holds copyright over software coded or influenced by AI.

AI’s place in software development

Generative AI has revolutionised the way software engineers approach new projects. Rather than relying solely on human expertise, AI streamlines the process of writing new code by following prompts from the developer, creating convenient scaffolds, suggesting more efficient processes and foreseeing errors in code.

Industry leaders hold that AI isn’t simply a magic machine which churns out work unprompted. New Jesey-based, Ness Digital Engineering, found that senior developers had a 48% rise in productivity when using AI tools, while junior engineers hardly benefited at all due to a lacking in the more basic skills needed to use the tools effectively². While generative AI might produce unique outputs, this is dependent on the creator’s ability to prompt with the right questions and use it to supplement original work.

Recognition under the Copyright Act

As it stands, AI itself doesn’t have any rights under copyright law. Thus, assuming the output is recognised as copyright material, the human contributor would need to be able to establish ownership. So, how is such recognition established?

Under the Copyright Act 1968 (Cth), where copyright in a work will only subsist where the work (such as the source code of software) is “original”. Under copyright law, original does not mean never done before – instead it looks at the skill, labour and judgement applied to the creation of the work by a human author.

Establishing originality in the context of AI generation hasn’t yet been tested, however litigation has ensued in the past on just how ‘original’ such works need to be.

When the High Court navigated this issue in Nine Network Australia Pty Ltd v IceTV Pty Ltd [2007] HCA 14, they held that for a work to be protected, it must have been of an “independently intellectual effort”, or at least “sufficient effort of a literary nature”. This was in the context of a dispute bought against IceTV, which was using programming data published by the Nine Network. IceTV were found not to have infringed on Nine’s copyright, largely due to the very limited degree of skill and labour required in essentially transcribing program titles with their start and finishing times. Even the names of the programs were at the discretion of the show producers, rather than the Nine Network. Thus, it couldn’t feasibly be stated that Nine’s presentation of the programming schedule was an original literary work, despite all the input being completed by Nine employees.

So how might creators evidence their skill and labour where an output is assisted by non-human creators?

Originality and AI

In Telstra Corporation Limited v Phone Directories Company Pty Ltd [2010] FCAFC 149, the Federal Court had to rule on whether yellow and white pages data published by Telstra were sufficiently original to be copyrighted. The clincher? In the words of CJ Keane, “the directories were compiled, not by the individuals engaged to facilitate the process, but by a computerised process of storing, selecting, ordering and arranging the data to produce the directories”.

The Court held that databases created using such processes didn’t qualify for copyright protection. Like in Nine v IceTV, the court ruled again based on a lack of originality. However, this time it was purely due to minimal human effort.

Despite Telstra’s argument that the directories were original works due to the significant effort and expense that went into the collation of the data, the court held that the actual directories were produced by automated systems with little to no creative input from humans. In essence, the court found that there is an intrinsic link between human authorship and originality, and that the purpose of the s32(1) was to protect the creative pursuits of individuals.

Implications for software developers

As discussed, even outputs which are informed by huge amounts of privately sourced data might be subject to no protection, where no human modification of the output occurs.

At the least, the courts have been unwaveringly reluctant to award any recognition for work completed that didn’t include some “sweat on the brow” which stemmed from the actual creation of the final product. Currently, it’s not clear whether complex prompting would constitute sufficient skill and labour.

Currently the development industry holds that AI only works as a supplement to original work, however the fine line of “independent intellectual effort” may become increasingly blurred as the technology advances.

While developers work assisted by AI, some best practices should be observed to ensure and evidence that a sufficient degree of human skill and labour has been met. Human review and auditing are fundamentally important to developer accountability and understanding of the code being used, which in turn will assist down the track where bugs arise. Businesses should also brush up on their recordkeeping and document traceability to evidence work authored or modified by humans. Software development tools and logs which can evidence or require human edits or modification to code before it is deployed in proprietary software will also be helpful.

Businesses looking to engage software developers should ensure appropriate obligations are included in the software development agreements, requiring the work produced to be original and the product of the skill and labour of a human developer. Otherwise, you may be paying a developer to produce code written almost entirely with AI and no copyright capable of being assigned to the business.

AI as a legal person

Currently, and for the foreseeable future, only ‘legal persons’ are protected under the Copyright Act. However, as AI evolves from its ‘generative’ phase to what industry predicts to be an ‘interactive’ phase, and beyond, some more speculative advocates for AI suggest it may one day be credited with “variants of legal personhood attributes” ³.Thus, paving the way for co-authoring arrangements under Part 3, Div 9 of the Copyright Act. However, this is purely conjecture, at least until AI bots become reproducing and conscious beings of their own!

Nothing short of developed case law, and regulatory reform is likely to provide solid guidance on the matter of IP protection for AI-augmented code. However, if the precedent in Telstra v Phone Directories is to be followed, in the words of CJ Keane, it “may give rise to a perception of injustice” for creators who prompt with their own data.