AI Acceptable Use

Vaario helps people compare versions of a document, see what changed, and read summaries and answers about it. It is an aid to your own review, not a replacement for your judgement, and it does not give legal advice (Terms of Service, §3). This policy adds the rules for the AI part. If it conflicts with the Terms, the Terms prevail.

When you upload a document or ask a question, Vaario sends the text it needs for that request to a language model (Claude, run through Amazon Bedrock in EU regions). The model works on it in memory and returns the result to your screen. Your content is not used to train any model. The privacy policy has the detail, including how long your documents are kept: the period you choose, 30 days without use by default.

Everything the AI writes is marked as coming from the AI: the chat is called “Ask the AI”, and edits it proposes carry an “AI edit” label.

The AI works by probability. It can:

• misread a complex clause;
• leave out something that is in the document;
• state facts, figures or conclusions that are not in the document at all;
• miss the context of your business or case, because it sees only the text it is given.

A person must review the AI's output before it is used for anything that matters, such as signing a contract, advising a client or sending a report. At the least:

1. Check the result against the original document for omissions and inventions.
2. Check that it fits your jurisdiction, your sector and the facts of your case.
3. Take responsibility for what you use. If you adopt an AI result in your work, you answer for its accuracy.

• Forge or deceive. Do not use it to make or alter documents meant to mislead others: fake signatures, altered evidence, or documents made to look as if an authority issued them.
• Decide about people without a human. Do not use its output to make decisions about a person that have legal effect or affect them in a similarly significant way (hiring, credit, profiling) without a meaningful review by a human.
• Attack the AI. Do not try to get round its safeguards, inject instructions into a document or prompt to change its behaviour, or extract its internal instructions.
• Train other AI. Do not use Vaario's output to train, tune, distil or benchmark another AI system.

On a team account the organisation that holds the account makes sure the people using Vaario understand that the AI can be wrong and follow the duty to check above before they use a result in the organisation's work.

Depending on how serious and how deliberate it was, we may warn you, limit the analysis features, suspend the account, or end the contract. We may suspend without notice only where there is a serious security risk, clearly unlawful activity or a lawful demand from an authority.

If the AI behaves oddly, gives an answer you think is harmful, or you suspect someone is misusing it, write to privacy@vaario.app. A technical vulnerability goes through the vulnerability disclosure policy.