AI Intensify: Practical AI for small business owners, explained plainly, with the real costs.

© 2026 AI Intensify

AI for Business Owners 5 min read

Who Owns the Work Now? AI Legal Questions for Agencies

Ownership of AI-assisted work, training data lawsuits, client data and EU rules: the AI legal questions agencies face, plus a one-page contract review.

AI-generated article. This article was researched and drafted using AI tools and published automatically, and its featured image was generated by AI. Facts are drawn from the sources cited in the text. How AI Intensify uses AI
Abstract illustration about AI legal questions: stack of translucent paper layers, some drawn by a steady hand and some by a mechanical arm, held together by a single brass paperclip under soft side l

Many agency contracts still promise that every deliverable is original work the client will fully own. If AI made part of that work, the promise may be hard to keep, because in the US a prompt alone does not make anyone an author. Here are the five legal questions that matter and what to check in your contracts this month.

The opportunity is easy to see: faster drafts, more options for clients, smaller teams doing bigger jobs. The cost is harder to see, because it arrives as uncertainty rather than an invoice, and most small agencies have no in-house lawyer to sort it out. What follows is general information drawn from public sources, not legal advice. For a specific contract or dispute, a qualified lawyer in the relevant country is the right call.

Who owns what the tools produce

The clearest answer so far comes from the United States. In January 2025 the U.S. Copyright Office published the copyrightability part of its AI report, and, as summarized by Mayer Brown, it treats prompts alone as instructions that convey unprotectable ideas, not as authorship. On March 2, 2026, the Supreme Court declined to hear Thaler v. Perlmutter, which leaves standing the appeals court ruling that a work needs a human author to be copyrighted.

What remains open is the middle ground. Mayer Brown notes there is still no bright-line test for how much human contribution is enough, and that Allen v. Perlmutter, a challenge involving an image refined through more than 600 prompts, may help clarify it. For an agency, the practical consequence is uncomfortable. A logo or campaign line produced mostly by a tool may carry thin or no copyright protection, which matters when the contract promises to assign ownership of it. Mayer Brown’s suggestion is simple: document the human creative input. Sketches, edits, selection decisions and revision notes are the record that shows a person shaped the result.

Training data fights, and who carries the risk

The other ownership question runs backward, to how the models were built. In July 2026 a federal court granted final approval to the $1.5 billion settlement in Bartz v. Anthropic, with roughly $3,000 per covered book, according to the Authors Guild. The release is narrow. It covers past acquisition and copying of the listed works, and it does not release claims about AI outputs or future conduct.

For an agency that only uses these tools, the more immediate issue is contractual: if a client is accused of infringing because of something an AI tool produced, who pays? Some vendors offer indemnities, but they come with conditions. Microsoft’s Customer Copyright Commitment, for example, requires Azure OpenAI customers to keep specific safeguards in place, such as a system instruction against reproducing protected material and documented testing, and it excludes output from outside models connected to Copilot Studio. An indemnity is only as good as its fine print. Most consumer subscriptions offer nothing comparable.

Client data and the promises made about it

Pasting a client’s customer list, sales figures or unreleased product plans into a consumer chatbot is probably the most common exposure, and the most avoidable. The Federal Trade Commission has warned that AI providers which quietly use customer data for training risk enforcement, and that promises made in marketing pages or terms of service can carry the same weight as a formal privacy policy. The same logic applies one step down the chain. An agency’s confidentiality clause is a promise too.

Agencies also write the claims their clients publish. When the FTC launched Operation AI Comply in September 2024, its first cases included DoNotPay, which agreed to pay $193,000 over its “robot lawyer” claims. The agency’s line was blunt: there is no AI exemption from existing consumer protection law.

The EU rules that moved, and the ones that did not

Agencies with European clients got partial relief this year. The Digital Omnibus on AI entered into force on July 27, 2026, according to Hunton Andrews Kurth, pushing the main high-risk obligations to December 2027 and, for AI built into regulated products, to August 2028. Transparency duties for AI-generated content were not delayed. Gibson Dunn notes they apply from August 2, 2026, with a grace period to December 2, 2026 only for the marking requirement on systems already on the market. For most marketing work, that transparency layer is the part that bites first. A plain-language walkthrough is in the post on the transparency rules now in force.

Keep this for every key deliverable

  • Early sketches or notes
  • Which AI outputs were picked, and which were rejected
  • Edits made by a person
  • Revision notes and dates

Working through the AI legal questions, one page at a time

None of this requires a legal department. It requires reading four or five documents the agency already has, with these questions in hand.

Question What the sources say What to check this month
Can the agency assign ownership of AI-assisted work? Prompts alone are not authorship in the U.S.; the line for mixed work is unsettled (Mayer Brown). Ownership and originality clauses in client contracts. Is there a record of human edits on key deliverables?
Who pays if output infringes? Vendor indemnities exist but carry conditions (Microsoft Learn). Which tools the team uses, which plans include an indemnity, and whether its conditions are actually met.
Where does client data go? Data promises in terms and marketing are enforceable (FTC). Each tool’s data-use settings, and whether the agency’s confidentiality clause allows that tool at all.
Are AI claims in client copy defensible? No AI exemption from consumer protection law (FTC). Any “AI-powered” claim the agency has written: can the client back it up?
Does content need an AI label? EU transparency duties apply from August 2026 (Gibson Dunn). Which clients sell into the EU, and who is responsible for disclosure in the contract.

The tempting response is to try to fix everything at once. It is more useful to pick the single contract that brings in the most revenue and read its ownership clause against the first row of the table. More context on where courts and regulators stand is in the overview of AI copyright and the guide to customer data risk in AI tools, and related posts sit under AI for Business Owners.

The harder question is the one clients will eventually ask out loud: how much of this was made by a person? An agency that can answer it with a record, not a guess, is in a far better position than one that cannot.

Up next

7 AI Automation Tools for a Streamlined Workflow, Priced for Small Business

Seven AI automation tools checked against vendor pricing pages in November 2026, with free options, entry prices and the skill each needs for small…

Keep reading →