AI Image Generation for Business: Licensing and Copyright Basics

You can generally use images from commercial AI image tools in marketing and products, but three separate questions decide how safe that is: what the vendor's terms say you may do, whether the vendor will defend you if someone claims infringement, and whether you can claim copyright in the image yourself. The answers differ by product and plan, and none of them is guaranteed by the others. This article explains each question using vendor documentation and the US Copyright Office's published position, and ends with a checklist. It is general information, not legal advice, and laws differ by country.

Who this is for and what you will have

This is for marketing leads, designers, small business owners and IT or procurement staff who approve AI tools. By the end you will be able to separate the ownership, indemnity and copyrightability questions, know what Microsoft and Adobe publicly state, and apply a pre-publication checklist with a simple record-keeping routine.

Question 1: What do the vendor's terms allow?

Start with the terms for the exact product and plan you use, because consumer, business and enterprise offerings often differ. Look for:

• Whether commercial use of outputs is permitted.
• Who the terms say owns the output, or whether the vendor disclaims ownership.
• Restrictions on content types, such as real people, brands or sensitive subjects.
• Whether your prompts and outputs are used for training or review.
• Any attribution or watermarking requirements.

Microsoft's Copilot privacy documentation states that Microsoft does not claim ownership of the output of the service. It also states that Microsoft does not make a determination on whether a customer's output is copyright protected or enforceable against other users, because generative AI systems may produce similar responses to similar prompts from multiple customers, so multiple customers may have or claim rights in the same or substantially similar content. That sentence is the key to the whole topic: absence of a vendor ownership claim is not the same as you holding exclusive rights.

Question 2: Will the vendor stand behind the output?

Some vendors offer an indemnity, which is a contractual promise to defend you if a third party claims the output infringes their rights. Every indemnity has conditions, so read them.

Microsoft

Microsoft's Copilot privacy documentation says that if a third party sues a commercial customer for copyright infringement for using Microsoft's Copilots or the output they generate, Microsoft will defend the customer and pay the amount of any adverse judgments or settlements, as long as the customer used the guardrails and content filters built into the products.

For Azure OpenAI, Microsoft publishes a page of required mitigations for the Customer Copyright Commitment. It states that the commitment is a provision in the Microsoft Product Terms describing Microsoft's obligation to defend customers against certain third-party intellectual property claims relating to output content, and that the customer must also have implemented the mitigations required by the Azure OpenAI documentation in the offering that delivered the output. For Azure OpenAI, the universal required mitigations are:

• A metaprompt in the customer's offering directing the model to prevent copyright infringement in its output, for example the "Protected Material - Text" sample system message component.
• A testing and evaluation report, produced by the customer, showing the offering was tested for the output of third-party content, with the report retained and provided to Microsoft in the event of a claim.

The page lists additional per-use-case mitigations for text and code generation, such as protected material detection, and states that for image generation models, OpenAI transcription models and all other use cases there are no additional requirements beyond the universal ones. It also says the page is not an exhaustive list of what you must do to use Azure OpenAI responsibly, and that Azure OpenAI customers must comply with the Code of Conduct at all times. If you build your own image feature on Azure OpenAI, treat the metaprompt and the test report as engineering tasks with an owner and a date, because you will need to show them in a claim.

Adobe

Adobe states that Firefly was trained on Adobe Stock images, public domain content and other openly licensed or non-copyright material. Adobe's enterprise page indicates that customers can purchase an entitlement that includes contractual IP indemnification for select Firefly outputs, subject to the applicable terms, conditions and exclusions, and press coverage of the programme says it covers the generated output and not elements the user adds. Which plans include the indemnity depends on your contract, so ask your account representative for the written terms rather than relying on marketing pages or third-party summaries. Programme terms have changed since the 2023 launch, so check the current version.

What to take from this

• Indemnities are conditional. Skipping a required guardrail or mitigation can remove coverage.
• They typically concern third-party infringement claims, not whether you own the image.
• They usually apply to the output as generated. Material you add or edit, such as a trademarked character or someone's photo, is your responsibility.
• Procurement should file the signed terms and the plan or entitlement that triggers coverage.

Question 3: Can you claim copyright in the image?

This is separate from the other two. The US Copyright Office published Part 2 of its report on copyright and artificial intelligence, on copyrightability, in January 2025. According to summaries from the Copyright Alliance, IPWatchdog and law firms, the Office concluded that:

• Copyright protects human authorship, and output wholly generated by AI is not protected.
• "Given current generally available technology, prompts alone do not provide sufficient human control to make users of an AI system the authors of the output."
• Protection can extend to human contributions, such as minimally creative selection, coordination or arrangement of AI-generated material, or minimally creative modifications of it, and to human-authored material that remains perceptible in the output.
• No new special right for AI-generated output was needed at that time.

These points come from secondary summaries of the report. Read the report on the Copyright Office website before relying on specifics, and remember it describes US law only. Other countries take different approaches, so if you sell or publish outside the US, check local rules.

The practical consequences for a business:

• A purely generated image may be copied by others without a copyright claim from you.
• Images that you substantially edit, combine or arrange with your own original work are a stronger candidate for protection in the human-made parts.
• Registration processes may ask you to disclose AI-generated content, so keep records of what was generated and what was human work.
• For assets that must be exclusive, such as logos, brand mascots and flagship campaign art, use human-created work or commission a designer and use AI only for ideas.

Risks that apply regardless of vendor

• Resemblance to existing work. Models can produce output that looks like a protected work or character. Reverse image search important assets before publication.
• Real people. Using someone's likeness can raise rights of publicity and privacy issues, and some tools restrict it.
• Trademarks and logos. An output may contain recognisable brand elements. Remove them, and run clearance for anything you will use as a mark.
• Names of living artists or brands in prompts. Avoid them. They increase the risk of a derivative result and may breach the tool's terms.
• Sensitive uses. Medical, political or news-style images can mislead. Label synthetic images where context could confuse viewers.
• Customer or employee data in prompts. Do not upload photos of people or confidential product images unless the tool's terms and your policies allow it.

Step-by-step: a pre-publication routine

1. Confirm the tool is approved and that you are on the plan your policy requires, not a personal account.
2. Read the terms and indemnity conditions for that plan and store a copy.
3. Write a prompt that describes the subject and style generically. Do not name living artists, brands or characters.
4. Generate several options and choose one yourself. Selection is a human decision you can record.
5. Edit the image. Original human work on the output, such as compositing, retouching or adding your own artwork, strengthens your position on the human-made parts.
6. Run a similarity check with a reverse image search for high-visibility uses.
7. Check for logos, text, faces and trademarks in the final image.
8. Record the details. Tool, plan, date, prompt, settings, the original output, the edited version and who approved it.
9. Add disclosure where appropriate, for example a note that an image is AI-generated, particularly in editorial or product-information contexts.
10. Escalate high-value or exclusive assets to a lawyer before launch.

A simple record format:

Asset: spring-campaign-hero.png
Tool and plan: <tool name>, <business plan>
Date generated: 2026-10-25
Prompt: <full prompt text>
Original output file: <path>
Human edits: <what was changed or added, by whom>
Similarity check: <done, date, result>
Approved by: <name, date>

Troubleshooting

The image looks like a known character or brand. Do not use it. Regenerate with a more generic prompt, or commission original artwork.

Legal asks who owns it and the vendor documentation is unclear. Ask the vendor for the written terms for your plan, and tell legal that vendor non-ownership and copyright protection are different things.

You built an image feature on Azure OpenAI and want coverage under the Customer Copyright Commitment. Implement and document the universal mitigations, and keep the evaluation report where you can find it. See also fixing Azure OpenAI content filter errors for how guardrails behave when they trigger.

A team member used a personal account for company images. Move the work to an approved business plan, and review the personal account's terms, because commercial rights and protections may differ.

Checklist

• Use approved tools on business plans, with terms and indemnity conditions filed.
• Treat ownership, indemnity and copyrightability as three different questions.
• Keep prompts generic: no living artists, brands, characters or real people.
• Add human creative work where you want to claim rights.
• Check for lookalikes, logos and trademarks before publication.
• Keep a record of tool, plan, prompt, edits and approval.
• Use human-made work for logos and other assets that must be exclusive.
• Get legal advice for high-value or high-risk uses.

References

• Microsoft Learn, Customer Copyright Commitment Required Mitigations: https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/customer-copyright-commitment
• Microsoft Learn, Data, Privacy, and Security for Microsoft Copilot: https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy
• Microsoft, Copilot Copyright Commitment announcement: https://blogs.microsoft.com/on-the-issues/2023/09/07/copilot-copyright-commitment/
• Adobe, Firefly for enterprise: https://www.adobe.com/ie/products/firefly/enterprise.html
• US Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability: https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf
• Copyright Alliance, Summary of the USCO report, Part 2: https://copyrightalliance.org/ai-report-part-2-copyrightability/