EssayBy The Talk AI Tools editorsNovember 5, 2026

Who owns what the model writes

Every assistant vendor has a sentence about who owns the output, and no two of them use the same verb. One assigns, one declines to claim, one declines to assert. The difference is not decoration.

All three sets of terms were read against the live page markup on October 25, 2026. The question that matters to a small company is narrower than ownership: can you sell what came out, and who carries the risk if it belonged to somebody else.

GitHub Copilot's public homepage, captured automatically on October 14, 2026. Not a logged-in account; the page may have changed since.

Anthropic assigns its interest, and qualifies the assignment in the same breath

The commercial terms put the conclusion first. Anthropic agrees that the customer "retains all rights to its Inputs" and "owns its Outputs", and "disclaims any rights it receives to the Customer Content under these Terms".

Then comes the operative clause. "Subject to Customer's compliance with these Terms, Anthropic hereby assigns to Customer its right, title and interest (if any) in and to Outputs."

Read the parenthesis twice. The assignment covers whatever interest the vendor holds, and the terms decline to promise that any such interest exists.

The consumer terms use the same shape for individuals. They say the company assigns "all of our right, title, and interest" in Outputs, qualified in the same sentence by the words "if any".

Those two words are the honest part of the clause. A vendor can hand over its own rights; it cannot manufacture a copyright in text that a machine produced, and it does not claim to.

The definitions are worth keeping. "Inputs" are submissions to the service and "Outputs" are responses generated to those inputs, and together they form "Customer Content", which the commercial terms also treat as the customer's confidential information.

GitHub does not claim ownership, which is not the same as granting it

GitHub's Terms of Service, effective April 27, 2026, open the AI section with a flat negative. "GitHub does not claim ownership of your Input or Output."

That sentence resolves one question and leaves the commercial one open. Nothing in it says the output is yours; it says the vendor is not asserting a competing claim.

The next sentence explains why the distinction is live. "Output may contain material that resembles code or content in the model's training data or that is subject to third-party copyrights or open source license terms."

And the duty lands on the user. "You are responsible for determining whether your use of Output requires a third-party license and for complying with any such license."

For a small team shipping a product, that is the whole clause. The licence audit on generated code is yours to run, and the terms say so in plain words.

GitHub also separates ownership from training. The data-use licence in the same section applies "only to individual licenses", and an opt-out applies "from the effective date of your opt-out going forward" rather than retroactively.

Google classifies the output as your data, then declines to assert rights in it

Google Cloud's service specific terms take a third route and start with classification. "Generated Output is Customer Data."

The ownership line follows. "As between Customer and Google, Google does not assert any ownership rights in any new intellectual property created in the Generated Output."

Note the hedge that both Google and GitHub use and Anthropic does not. The phrase "as between Customer and Google" settles the two-party question and says nothing about a third party with a prior claim.

Google then adds the warning that belongs next to any ownership clause. Generative services "may provide inaccurate or offensive Generated Output", and may "produce the same or similar Generated Output for multiple customers".

That last clause is the one small companies underrate. If a competitor's prompt can return your tagline, exclusivity was never part of the deal, whatever the ownership sentence says.

The indemnity tells you what ownership is actually worth

Ownership on paper matters less than who defends the claim. Anthropic's commercial terms promise to defend a customer against a third-party allegation that paid use of the services, or outputs from it, "violates any third-party intellectual property right".

The exclusions then narrow it, and two matter for anything commercial. The defence excludes "the practice of a patented invention contained in an Output". It also excludes "an alleged violation of trademark based on use of an Output in trade or commerce".

Google's version covers more ground and carves out more carefully. Its indemnity reaches "an unmodified Generated Output" from a listed indemnified service, and separately reaches allegations about the training data used to build the model.

The carve-outs are specific. The output indemnity drops away where the customer "knew or should have known" the output was likely infringing, where safety filters or source citations were disabled or circumvented, or where use continues after notice from a rightsholder.

There is also a price condition that is easy to miss. A "Generative AI Indemnified Service" is one on the published list "where the use of such Service or feature is not provided to Customer free of charge".

GitHub takes the opposite position and states it plainly. "Your indemnity obligations in Section Q apply to your use of AI Features and Output, including claims arising from Output you incorporate into your products or services."

Four sentences to find before you ship generated work

The ownership verb. Assigns, does not claim, or does not assert. Only the first transfers anything.
The qualifier. "If any" and "as between Customer and Google" both leave the third-party question open.
The licence duty. GitHub puts the third-party licence check on the user, in one sentence.
The indemnity, and its price. Google's output indemnity applies to paid, listed services only; GitHub's indemnity runs the other way.

What follows in practice

Treat the ownership clause as a release rather than a grant. It tells you the vendor will not sue you, and it stops there.

Then read the indemnity as the real product. One vendor defends unmodified output from paid services, one defends paid use against intellectual property claims with patents and trademarks carved out, and one asks you to defend it.

If generated material is going into something you sell, keep a record of which tool produced it and under which plan. The paid-tier condition in Google's definition makes that record part of the legal position, not just housekeeping.

The adjacent question of what the vendor does with your side of the exchange is examined in who reads what you type and the default that trains on your work.

Portability is the other half of ownership, and it rarely survives contact with a pricing page; that argument is in the claim that you can always switch.

Terms differ more than feature grids do, which is part of the case made in the tool is not the workflow.

Subject pages for the three assistants whose terms are quoted or implicated here: Claude, ChatGPT and GitHub Copilot.

Plan-level detail sits in our primer on AI assistant pricing for small teams, and the shortlist in best AI coding assistants for small teams.

Ownership of model output is not a single fact that a contract can settle. It is a release from the vendor, a duty handed to you, and an indemnity whose edges are drawn with unusual care.

Sources, read against the live page markup on October 25, 2026: Anthropic Commercial Terms of Service, effective June 17, 2025, and Consumer Terms of Service, effective October 8, 2025; GitHub Terms of Service, effective April 27, 2026; Google Cloud Service Specific Terms.