AI Transparency Notice

This notice supports the transparency and AI-literacy expectations of the EU AI Act (Regulation (EU) 2024/1689). It is written for users, in plain language. It sits alongside our AI Disclaimer, Terms and Privacy Policy.

1. You are interacting with an AI system

Every analysis, rewrite, draft, follow-up chat reply and rehearsal counterpart in Yurify is generated by an AI system. No human reviews your content before it is analysed. In Rehearsal mode the "other person" is an AI simulation, not a real human being, and is labelled as such in the interface.

2. What the system does

Yurify takes text (or audio/images you convert to text) that you submit, and returns interpretive output: an estimate of tone and escalation risk, possible readings of the message, suggested rewrites, and coaching questions. It is a writing and reflection aid. It does not decide anything about you or anyone else.

3. Classification under the EU AI Act

Yurify is not a high-risk AI system under Annex III: it is not used for recruitment, worker management, education, credit, essential services, migration or law enforcement, and those uses are contractually prohibited in our Terms. Yurify makes no claim of certification, conformity assessment, or CE marking, and none is required for a system of this type.

4. Workplace use restriction (Art. 5(1)(f))

The EU AI Act prohibits AI systems used to infer the emotions of a natural person in the workplace or in education. Yurify is a personal tool for an individual reflecting on their own communication. It must not be deployed by an employer, manager, HR function or educational institution to read, score, monitor or assess employees, candidates or students. When you select a work context, Yurify is instructed to analyse how your wording is likely to read at work, rather than to infer the inner emotional state of a named colleague, and the output must not be used as input to any employment decision.

5. Models behind the product

Requests are routed through a managed AI gateway. Providers are instructed not to train on your content to the extent their enterprise terms allow. The current sub-processor list and transfer safeguards are in our sub-processor page.

6. Known limits and failure modes

• Outputs are statistical predictions. They can be wrong, invented, or internally inconsistent.
• Two runs on the same text can differ.
• Quality varies by language, dialect, slang, and writing style.
• Training data carries cultural, gender and age bias; readings of tone may reflect it.
• Short or context-free messages produce weaker, more speculative readings.
• Terms such as "emotion", "intent", "bias" or "real need" are interpretive labels, not clinical findings.

7. How to read the numbers

Scores (0–10) express how strongly a signal appears in the text, not a measurement of a person. In the cognitive-bias panel, the percentage is the model's confidence that a reasoning pattern is present in the wording, and "sev X/10" is how much that pattern is likely to distort the exchange. Both are estimates produced by the model itself, not validated psychometrics.

8. Human oversight, contesting and reporting

You remain the decision-maker: nothing is sent, filed or shared on your behalf. If an output is wrong or harmful, use "Report this analysis" under any result — choose "AI output was wrong or harmful" — or write to info@yurify.co. Reports reach a human, and we use them to correct prompts and guardrails. You can also delete your analyses and account at any time from settings.

9. Content generated by AI

Drafts, rewrites and simulated replies are AI-generated content. When you copy, export or share them, they are labelled as AI-generated. If you forward them onward, please keep that context — presenting AI-generated text as another person's own words can mislead the reader.

10. Contact

Questions about this notice or about how Yurify's AI works: info@yurify.co.