AI output disclaimer

An output is a piece of text, code, image, or structured data produced by Ultron in response to a prompt. The platform composes a foundation model with your data, your connected tools, and the skill prompt you triggered. The output reflects all of those inputs plus the probabilistic behaviour of the underlying model. It is a draft, prepared in seconds, that benefits from a human review before it is used.

An output is not a guarantee of accuracy. It is not the truth as confirmed by authoritative sources. It is not professional advice. It is not a finished work product unless you, the human, agree that it is. The model does not know what it does not know; it produces fluent text whether or not the underlying claim is true.

You are responsible for what you do with an output. The platform supports that responsibility by surfacing the tools the model used, the citations it produced where it produced them, and an approval step for actions that leave the platform. Verification is not optional for anything that matters. Specifically:

• If the output cites a source, click the source. The platform does not guarantee that citations are accurate or that the underlying source supports the claim.
• If the output makes a specific factual claim (a person's title, a company's funding, a regulation), confirm it against an authoritative source before relying on it.
• If the output is code, read it. Tests pass on what you wrote, not on what you meant.
• If the output describes a person, read the section on output about identifiable people below.

Ultron does not provide medical, legal, financial, tax, accounting, or other professional advice. Outputs that read like advice are not a substitute for a qualified professional. Do not act on an Ultron output in any of those domains without consulting a licensed practitioner who can apply the law and the facts to your specific situation.

Some decisions can change a person's life. Hiring, firing, credit, insurance, immigration status, custody, criminal justice screening, access to housing, access to healthcare. Ultron is not designed to be the final decision-maker in any of those domains. If you build a workflow that touches them, a qualified human must read the output and own the decision before any consequence reaches the affected person.

Large language models can produce text that reads authoritative even when the underlying facts are wrong or invented. This is called hallucination. The risk is highest for specific numbers, dates, quotations, legal citations, and statements about niche or recent topics. Skills that rely on web search reduce the risk because they ground the model in retrieved text, but they do not eliminate it. The model can still misread a source or extrapolate beyond what the source supports.

Output that describes, profiles, or makes claims about an identifiable individual is at higher risk of harm. A false statement about a real person can be defamatory. An inference about a sensitive attribute (health, religion, sexual orientation, political views) can violate data protection law even if it is not published. Treat outputs that name or imply a person as drafts that require explicit verification and, where the statement is material, the consent of the person involved.

Do not use Ultron to generate explicit or sexual content involving identifiable individuals without their consent. Do not use it to fabricate quotes attributed to real people. Do not use it to make claims about a person's criminal history, medical status, or finances that you cannot substantiate.

Outputs from generative models are not unique. The same prompt can produce similar or identical outputs for different users. Outputs may reflect patterns learned from the model's training data, which can include public material that resembles a particular author's style. You are responsible for not passing off the work of another author as your own and for any copyright clearance required for your use.

Some jurisdictions and platforms require AI-generated content to be labelled, especially when it is shared publicly, used in advertising, or attributed to a real person. The EU AI Act, the Digital Services Act, and several national rules apply. You are responsible for complying with the labelling rules that apply to your distribution channel and audience.

To the extent we have any rights in the outputs you generate, we assign those rights to you. You are free to use, modify, and distribute the outputs subject to the law, to the AUP, and to the rights of third parties whose work or data may be reflected in the output. Ownership of an output does not extinguish the obligation to review it before publishing it.