Copyright is the legal protection that gives authors exclusive rights to reproduce, distribute, perform and adapt their original creative works. AI raises copyright questions on both ends of the pipeline: whether training on copyrighted material is lawful (covered in the fair-use entry), and whether AI-generated outputs themselves can be copyrighted. The second question matters enormously for businesses that ship AI-generated content commercially.
The US position as of 2026:
- Pure AI-generated works are not copyrightable — the US Copyright Office has held consistently since 2023 that copyright requires human authorship; pure AI output cannot be registered.
- Human-authored works incorporating AI elements can be copyrightable — but only the human-authored portions, and applicants must disclaim the AI-generated parts during registration.
- Selection and arrangement of AI-generated elements may be protectable as a compilation, even when the underlying elements are not.
- Prompts alone are generally not enough — writing a prompt does not constitute authorship of the resulting image or text under current US guidance.
The practical implications for US businesses in 2026:
- AI-generated marketing copy, images and video have reduced protection — competitors can theoretically copy them with less risk than copying human-created equivalents.
- Hybrid workflows preserve more rights — substantial human editing, curation and arrangement strengthens the copyright claim on the final work.
- Documentation matters — keep records of human contributions for any work you may want to enforce copyright on.
- Disclosure on registration — when registering with the US Copyright Office, AI involvement must be disclosed and the AI-generated portions disclaimed.
The international landscape:
- UK allows copyright in computer-generated works under section 9(3) of the CDPA, with the human who arranged for the creation as the author. A different model from the US.
- EU generally requires human authorship; the EU AI Act adds transparency obligations on AI-generated content.
- China has held in some cases that AI-generated images can be copyrighted when there is sufficient human creative input.
- Japan is generally permissive on AI training but the rights to AI outputs are still being clarified.
The provider terms of service add another layer:
- OpenAI, Anthropic, Google assign output ownership to the user (subject to terms) but do not warrant that outputs do not infringe third-party rights.
- Adobe Firefly — trained on licensed and public-domain content with explicit IP indemnification for commercial customers; a notable safer-harbor offering.
- Midjourney, Stable Diffusion — commercial use generally permitted under their terms but no infringement indemnification.
- GitHub Copilot Business / Enterprise — Microsoft offers IP indemnification for generated code.
- Many Microsoft and Google enterprise AI offerings — increasingly include some form of IP indemnification as a competitive feature.
For a US team shipping AI-generated content commercially in 2026, the operational posture is: assume the work has limited copyright protection, do meaningful human editing for anything you might want to defend, document the human contribution, use providers with IP indemnification for higher-risk content, and disclose AI involvement clearly when required. The legal landscape will continue to evolve; courts will eventually answer the questions current Copyright Office guidance only partly addresses, and the right time to revisit your posture is annually.
