Ownership vs. Output – Solving the AI IP Puzzle

With the 2026 Budget incentives lowering the cost of entry, SMEs are no longer just using AI to summarise emails—they are using it to generate marketing copy, write software code, and even design product prototypes. But as your “AI-enabled” business scales, a high-stakes legal question emerges: Who actually owns what the AI produces?

In Singapore’s intellectual property (IP) landscape, the lines between human creativity and machine output are being redrawn. For an SME, failing to secure these rights can lead to a “hollow” business where your most valuable digital assets aren’t actually yours to sell or protect.

The Authorship Void: Can a Machine Own Copyright?

The current position of the Intellectual Property Office of Singapore (IPOS) remains anchored in the principle that copyright requires a human author. This creates a significant “ownership gap” for SMEs. If you use a generative AI tool to create a logo or a blog post with a single-sentence prompt, the law may view the resulting work as lacking the “authorial contribution” necessary for copyright protection.

This means that a competitor could, in theory, scrape your AI-generated website content or use your AI-designed marketing visuals without infringing your copyright, because no copyright existed in the first place. To bridge this gap, SMEs must move beyond simple prompts. Human-AI collaboration—where your team significantly edits, arranges, or adds creative layers to the AI’s output—is the only reliable way to ensure the final product qualifies for protection under the Copyright Act 2021.

The “Accidental Plagiarism” Trap

While you worry about owning your output, you must also worry about whose work went into the AI. Most Large Language Models (LLMs) are trained on vast datasets of public information. There is a persistent risk that an AI-generated output could inadvertently mirror a protected work too closely, leading to claims of copyright infringement against your business.

Singapore’s Computational Data Analysis (CDA) exception allows for the use of copyright material for purposes like training AI models, but it does not protect the user if the output of that AI is a “substantial reproduction” of someone else’s work. As a business owner, you cannot use “the AI did it” as a legal defense.

The Trade Secret Strategy: Protecting Your “Secret Sauce”

Because copyright and patent protection for AI can be elusive, many savvy Singaporean SMEs are turning to Trade Secret law.

Under the Singapore IP Strategy 2030, trade secrets have become a vital tool for the digital economy. While you might not own the copyright to a generic AI response, you can protect:

• Custom Prompts: The specific, highly engineered prompts your team uses to get superior results.
• Fine-Tuning Datasets: The proprietary business data you used to train a local model.
• Model Weights: The specific internal parameters of a locally hosted model that give your AI its unique “edge.”

Unlike patents, trade secrets require no registration and never expire—provided you take “reasonable steps” to keep them secret. This includes using Non-Disclosure Agreements (NDAs) with employees and ensuring your AI vendor contracts don’t give the provider a license to use your proprietary inputs to train their global models.

Contracting for Certainty

When tapping into the Enterprise Innovation Scheme (EIS) to hire AI consultants or vendors, the contract is your primary shield. Ensure your agreements explicitly state that:

• All IP rights in the deliverables (including AI-assisted outputs) are assigned to your SME.
• The vendor provides an indemnity against any third-party IP infringement claims arising from the AI’s training data or output.

As we move into Article 4, we will look at how to manage the “human side” of the machine: ensuring your AI isn’t just productive and protected, but also fair and ethical.