AI & Content Policy
Who owns generated output, what provenance is recorded, where restrictions come from, why a person always approves, and what a model’s output is worth as fact.
Updated July 2026
This policy covers the part of Aiora Labs that generates things. It is written to be usable rather than defensive: what you own, what we record, what we will not let an agent do on its own, and what you should never assume about a model’s output.
Ownership of outputs
You own the outputs you generate, subject to the terms of the model provider that produced them. You also keep the rights in whatever you brought with you: your prompts, your uploads, your reference media, your brand material.
Where a provider imposes a restriction, for example on commercial use, that restriction is shown against the model at the point you choose it rather than buried in a policy page. You should not have to read this document to find out what you are allowed to do with something you just made.
What ownership does not settle
Owning an output is not the same as being clear to publish it. An output can be yours and still infringe someone else’s work, depict a real person without a basis, or breach a platform’s rules where you post it. Those are separate questions and this policy does not answer them for you.
Nothing here is legal advice, and whether a particular output is safe to use commercially depends on the output, the jurisdiction and the use.
Human accountability
Agents draft and propose; a person approves before anything is billed or published. This is a design position rather than a temporary limitation. An agent that can spend your money or post in your name unprompted is not a feature we want to build, so we have not built it.
You remain responsible for what you approve. An agent acting on your instruction is you acting, and approving something quickly is still approving it.
Spend is always confirmed
Every generation is quoted in credits before it starts and nothing begins until you confirm that number. There is no post-hoc billing, because there is no post-hoc billing path. The confirmation is what creates the run.
Agents inherit the same rule and cannot route around it. Where you set a spend threshold, anything above it waits for your review rather than proceeding and telling you afterwards.
Provenance
Generated media carries its origin: which model produced it, when, and from which project and prompt. That record is attached to the item rather than kept in a separate log, so it travels with the work between tools.
Provenance is there for your benefit as much as ours. Six months later, the question “how did I make this?” has an answer.
Labelling AI-assisted work
Where labelling AI-assisted content is required, by law or by the platform you are publishing to, we label it. We would rather over-disclose than have you discover a disclosure obligation after publishing.
Outside those cases, how you describe your own process is yours to decide. The Community Guidelines do ask you not to actively misrepresent it.
Model providers and restrictions
Generation runs on third-party models under their usage terms. Some content types are restricted by the provider rather than by us, and where they are, the restriction appears at the point of use rather than as an unexplained failure.
We describe these suppliers by what they do rather than by name. Which model ran your generation is recorded against the output; which company stands behind it is not something we advertise on a marketing page.
Models change
Models are added, updated, re-priced and withdrawn by their providers, and a model can behave differently after an update without us changing anything. Where a model is withdrawn, the work you already made with it is unaffected. It is yours and it stays in your library.
Price changes take effect at the next quote. A run already confirmed at a price runs at that price.
Accuracy
Generative models invent things. Output can be wrong, biased, or confidently false, and fluency is not a signal of correctness. A model is at its most convincing exactly where it has no idea.
Nothing produced here should be treated as fact without checking it, and that goes double for anything that looks like a citation, a statistic, a quotation, a name, a date or a legal or medical statement. Assume it needs verifying, because it does.
What we will not generate
The prohibitions in the Acceptable Use Policy apply to generation as much as to upload: nothing sexualising minors, nothing unlawful or facilitating serious harm, nothing hateful targeting people for who they are, and no depiction of a real identifiable person without consent or another lawful basis.
Attempting to work around a safety or identity guardrail, including by getting an agent to do it for you, is a breach of the terms in itself, whatever the output turns out to be.
Your private work stays private
Your private creations are yours. They are not used to train third-party foundation models and they are not sold. Prompts and media you submit go to the model provider needed to fulfil that specific run, under that provider’s terms, for that run.
Output is private to you by default. Sharing it is an action you take.
Reporting something
Report harmful, infringing or unsafe content through support. A person reviews it and we act: content removal, restricted features or a closed account, depending on what it is and whether it has happened before.
Copyright complaints have their own process in the Copyright & DMCA policy, which is the faster route for a rights claim.