Businesses rushing to embrace artificial intelligence risk giving away valuable intellectual property in products they thought they owned.
Melissa Miller, Digital & IP Partner at McCullough Robertson Lawyers, said businesses and start-ups were increasingly using generative AI to develop software, content and other products.
However, this was often happening without a full understanding of the potential impact on their intellectual property.
Ms Miller said the issue was becoming increasingly important as investors encouraged founders and businesses to use AI to improve efficiency.
She said copyright protection required sufficient human intellectual effort in the final work, creating a potentially significant problem when products were generated predominantly by AI.
“There are a lot of different motivations in creating products,” she said. “This includes a desire to commercialise them, sell them and attract investors or potential buyers.”
“But if the core value of what they’ve developed is depleted through the overuse of AI, it may not have any intellectual property protection.”
Protections available under copyright
Under Australian copyright law, copyright arises automatically rather than through registration but requires a human author to exercise independent intellectual effort in the material form of the work.
Entirely AI-generated works are unlikely to be protected by copyright, while the prospects improve as human creative involvement increases.
Ms Miller said this meant a business could successfully commercialise an AI-developed product only to encounter problems when an investor or potential buyer conducted due diligence on its intellectual property.
This was particularly the case when someone had a sellable idea, but they used AI to take it from an idea to a product.
Australian courts have not yet ruled on ownership of AI output, so the current position is based on cases involving computer-generated work. “If it’s fully generated by AI, not with any human involvement, or not with sufficient human involvement, then it is likely the output will not be protected by copyright, and that means others may be free to copy it and reproduce what you’re doing,” she said.
“In this situation you don’t have the exclusive rights that a copyright owner. would have”
Ms Miller said the solution was not necessarily to avoid AI, but to ensure humans remained meaningfully involved in creating the final product.
“Yes, you can use AI, but you need to be mindful of how much independent intellectual effort is directed to that final expression because ultimately copyright will attach to the expression in material form, not the idea,” she said.
The issue can be particularly significant in software development, where AI can now generate substantial amounts of code.
Ms Miller said original software code written by human programmers can be protected by copyright as a literary work, while businesses relying heavily on AI-generated code could find there was little or nothing they could own.
She said entrepreneurs should consider using AI output as a first draft rather than a final product.
This meant materially editing and restructuring it, making specific creative choices and retaining drafts, prompts and version histories as evidence of human intellectual effort.
Confidentiality risks in public AI platforms
A second major risk involved confidential information being entered into publicly available AI platforms.
Ms Miller warned that businesses could lose confidentiality simply by entering sensitive or confidential information into a public AI model, depending on the platform’s terms.
“You may lose confidentiality immediately by putting information into a public AI model,” she said.
“Putting important information that should be secret – like an invention that could be patentable – into a free or consumer version of AI, can result in a loss of confidentiality and put future patent rights at risk.”
She said businesses using higher-tier or enterprise AI products needed to examine their terms to understand how inputs and outputs could be used and whether information could be used for model training.
The risks extended beyond commercial information. Ms Miller said there was also risk around inadvertently waiving legal professional privilege.
Where copyright protection was uncertain, confidential information, trade secrets and know-how could become particularly important ways of protecting commercial value.
That required businesses to control access and disclosure through measures including confidentiality agreements, contracts and data security arrangements.
“If people are thinking only about IP as something you own, I would be encouraging people to think about it as something you control,” Ms Miller said.
“If you don’t have that because you’ve just put everything secret into public AI, then actually what you’re left with is potentially nothing at all.”
