More and more, I hear some version of the same question from business owners: “We made something valuable with the help of AI. Can we protect it?”
Sometimes the “something” is obvious, like marketing copy, a logo, a photo, a product description, a training guide, or software code. Sometimes it is less obvious but more important, like a pricing model, a customer segmentation strategy, an internal workflow, or a set of prompts that reliably produces good results for the business. Either way, the underlying hope is the same. If it is valuable and it feels like ours, we want a way to prevent other people from simply taking it.
What makes this tricky is that U.S. intellectual property law is built around human contribution. That shows up in different ways depending on the tool you are using, but the theme is consistent: copyright focuses on authorship, and patent law focuses on inventorship. When a machine is doing a meaningful portion of the expressive or inventive work, those systems start to feel less accommodating than many people expect.
Copyright is usually the first place people look because generative AI produces the kinds of things copyright is designed to cover: text, images, music, video, and code. The friction is that the U.S. Copyright Office has been very clear that copyright protects human-authored expression. If a system generates the expressive elements of a work, the Office does not treat that output as copyrightable in the usual way, because it does not view the output as the product of human authorship.
That does not mean “AI is involved” automatically kills copyright. A lot of modern creative work involves tools. The key question is whether the human contribution rises to the level of authorship. In practical terms, the Copyright Office has focused on what the person actually did. Did the person create original expression themselves, such as writing, editing, rewriting, selecting and arranging material in a creative way, or making meaningful modifications that reflect human creative choices? Or did the person mostly provide prompts and accept what the system returned?
Prompts matter, but not in the way most people assume. A prompt can be clever, detailed, and highly intentional, and you still might not have copyright in what the system produces, because the system is still making the actual expressive choices. The Office’s approach has generally been that if the system determines the words, the imagery, the structure, the style, or the composition, then the output is not human-authored in the way copyright requires.
Patent law raises a different set of issues. Patents protect inventions, not creative expression. In the AI context, that usually means technical methods, systems, and functional improvements. A slogan or an image is not a patent problem. A new data processing method might be.
The first issue is inventorship. The legal system expects inventors to be human. Courts have reinforced that point, and the USPTO has issued guidance that treats AI as a tool rather than an inventor. The analysis still turns on whether a natural person conceived the claimed invention under existing standards. AI can be part of the process, and it can make the process more powerful, but it does not get named as an inventor, and its role does not change the underlying test.
The second issue is strategic and practical. Patents require disclosure. You are trading secrecy for a government-backed right to exclude others. Sometimes that is exactly the right move, especially when your invention is likely to be independently developed by competitors, and you want a clear, enforceable right. Other times, particularly when the value lies in a fast-evolving workflow or in business know-how that you would not want to publish, the patent bargain starts to look less attractive.
This is where trade secret law often becomes the most business-friendly option for protecting AI-related value, at least right now.
Trade secrets are not about authorship or inventorship. They are about confidentiality and value. In general terms, a trade secret is information that is not generally known, that has economic value because it is not generally known, and that is subject to reasonable measures to keep it secret.
That definition turns out to fit a surprising amount of what businesses are actually building with AI.
Many companies are not trying to protect a single output in isolation. They are trying to protect a repeatable capability. They want to protect the workflow that consistently produces high-quality results, the prompt library that makes the model behave, the training data they curated, the evaluation methods they developed, the internal playbooks that keep things compliant, and the system that turns a general-purpose model into something specialized and reliable.
Those are not always good candidates for copyright or patents. They are often excellent candidates for trade secret protection, as long as the company treats them like secrets.
That last part matters. Trade secret protection is powerful, but it is not passive. You do not get it because information feels confidential. You get it because you took reasonable steps to keep it confidential. That usually means contracts, access controls, vendor discipline, internal policies, and a habit of asking, before you share something, whether sharing it destroys the value you are trying to protect. It also means thinking carefully about what you upload into third-party systems, including AI tools, and under what terms.
Trade secret law also has limits. It will not stop someone who independently develops the same idea without misappropriating it from you. It also will not help if what you are protecting is easy to reverse engineer from the public product. That is why, in some cases, patents still make sense. It is also why a good protection strategy is often layered rather than single-track.
In practice, many businesses will do best by thinking in categories. Some assets can be shaped into human-authored expression and protected by copyright, especially if you build a workflow that includes meaningful human writing, editing, and creative control. Some innovations can be protected with patents, if they are truly inventions and if the disclosure tradeoff is worth it. And much of the most valuable AI-related advantage, the internal know-how, the data, the process, the guardrails, and the system design, should be treated as a trade secret program, not an afterthought.
If you are using AI in your business, the most useful early step is often not filing something. It is taking inventory. What are we creating that has lasting value? What would actually hurt if a competitor got it? What do we currently expose without thinking? And are we behaving as if our workflows are assets, or as if they are disposable conveniences?
That shift in mindset is where I see the next year heading, toward clearer thinking about what kinds of value AI is creating inside businesses, and which legal tools fit that value best.