AI-generated assets in M&A: Who owns what?

Increasingly, companies are using AI tools to either create, or assist in the creation of, company assets such as source code, images, logos or marketing material and reports.

But who owns those assets?

This question has significant implications in a mergers and acquisitions (M&A) context. Below, we outline the intellectual property (IP) position in Australia and the practical risks for buyers and sellers.

IP in AI-generated assets

In Australia, the intellectual property treatment of AI generated assets is determined by the application of Australia's existing intellectual property laws including the Copyright Act 1968 (Cth) (Copyright Act).

Copyright is the main form of intellectual property used to protect materials such as documents, artwork and software. Under the Copyright Act, copyright subsists in literary works (e.g. computer programs, software code and books) and artistic works (e.g. images and diagrams) if they meet certain requirements. Relevantly, the works must:

• be created by a human author; and
• be "original".

Establishing "originality" in a work requires the author to demonstrate a sufficient degree of independent intellectual or creative effort by a human.

The precise degree, and nature, of the human effort which is required for AI generated works to obtain copyright protection is not specified in the Copyright Act nor has it yet been determined by the Australian courts. That said, based on the relevant provisions of the Copyright Act and the findings of courts in other jurisdictions, it is unlikely that copyright will subsist in works created by AI with little or no human input other than basic prompting.

If copyright does not subsist in a work, the work can be used or copied by third parties unless another restriction, such as a contractual confidentiality obligation applies. For example, if a company uses AI to generate images used in marketing material, a third party cannot be prevented from reproducing those images on the basis of copyright infringement.

Why this matters in M&A

From an M&A perspective the above position will impact:

• sellers - in terms of their ability to demonstrate ownership of AI generated assets (such as source code) and realise value by selling those assets to a potential buyer of their business; and
• buyers - in terms of the buyer's ability to exclusively exploit such assets.

Recommendations for buyers

To identify risk and avoid overvaluation, buyers should:

• Conduct thorough and specialised due diligence (DD) on the target. Such DD should consider:
  • the extent of AI tools used by the target and the purposes for which they are used;
  • the details of any AI generated code and other works/materials (including the degree of human input in their development);
  • contractual terms in place between the target and its AI tool provider(s); and
  • the risk(s) of the target's confidential information having been compromised in the processes of using relevant AI tools.
• Understand that AI generated assets are unlikely to attract copyright protection and considering whether other appropriate protections (such as confidentiality obligations) are in place. This will also impact the extent to which a buyer can rely on intellectual property warranties in relation to the assets.
• Consider from a commercial perspective to the value of any AI generated assets. This may involve a consideration of whether the value of the asset is tied to exclusivity in the asset (which will be negatively impacted by the works not being subject to copyright protection), or the ability to leverage such assets quickly to service the intended market.

Recommendations for sellers

To mitigate buyer concerns and preserve deal value, sellers should:

• Review and, where possible, negotiate AI provider contracts to address confidentiality and IP ownership.
• Train personnel to avoid uploading confidential information into open AI systems.
• Ensure meaningful human involvement in the creation of key assets.
• Document how assets were created, including the extent of human input.
• Carefully assess the scope of IP warranties that can be provided.

AI-generated assets can be commercially valuable - but assumptions about ownership may not withstand scrutiny. In M&A, clarity on authorship and protection is critical to value and risk allocation.