EU AI Act compliance

Yurify Pte. Ltd. is the provider of the AI system. We do not train foundation models; we build on general-purpose models supplied through a managed gateway, listed on our sub-processors page. Your organisation is the deployer: you decide who uses Yurify, for which conversations, and how output feeds into your own processes.

Yurify is a limited-risk AI system subject to the transparency duties in Article 50. It is not a high-risk system under Annex III, and it must not be used in a way that would make it one. Specifically, it is not used for recruitment or selection, for decisions on promotion, discipline or termination, for allocating work or monitoring performance, for evaluating people in education, or for access to essential services. Output is advisory preparation material for the person who wrote or received the message.

If your organisation intends to use Yurify in an employment-related decision process, that use is outside the intended purpose we declare, is prohibited by our terms of service, and would place high-risk and fundamental-rights-impact obligations on you as deployer that this product is not designed to support.

• Emotion inference in the workplace (Art. 5(1)(f)). For users in the EU, EEA, UK, or Switzerland, emotion inference is blocked at generation time, not merely hidden in the interface. The model is instructed not to produce emotional-state labels, scores, or synonyms for any person, so none are generated, displayed, exported, or stored. Where the region cannot be determined, we suppress by default for workplace relationships. Communication risk, ambiguity, evidence and suggested wording remain fully available.
• Biometric categorisation and identification (Art. 5(1)(g)-(h)). Yurify processes text. It performs no face, voice, or biometric analysis, and no inference of race, political opinion, trade union membership, religion, health, sex life, or sexual orientation.
• Social scoring (Art. 5(1)(c)). Yurify assigns no score, rank, level, or standing to any person, and produces no comparison between colleagues. The practice log is private to the individual and visible to no employer.
• Manipulation and exploitation of vulnerability (Art. 5(1)(a)-(b)). Output is generated under a safety layer that refuses coercive, deceptive, or retaliatory guidance and refuses to help conceal misconduct.
• Predictive assessment of individuals (Art. 5(1)(d)). No criminality, deception, or trustworthiness prediction of any kind.

• Users are told, before use, that they are interacting with an AI system.
• Every generated artefact, including PDF exports, carries a visible AI provenance marker.
• Chat, rehearsal, and copilot surfaces are labelled as AI-generated and never presented as a human.
• Model families in use, and the gateway that routes to them, are published on the sub-processors page.
• Known limits, including probabilistic output and variation across languages, are stated in the AI disclaimer.

Yurify makes no automated decision about any person. Nothing is sent, filed, or actioned by the system: it returns a draft and a reading for a person to accept, edit, or discard. Suggested wording is presented as a suggestion, and the user remains the author of anything they send. Under the GDPR, no output constitutes a solely automated decision within the meaning of Article 22.

Article 4 requires deployers to ensure a sufficient level of AI literacy among the people who use the system. To support that, brief your users on four points, all of which we also state in the product:

• Output is a statistical prediction, not a finding of fact, and two runs may differ.
• The system reads text you provide. It has no knowledge of the people involved and no access to your systems.
• Interpretive labels are shorthand for patterns in language, never a judgement about a person.
• The user stays accountable for what they send, and for checking it against what they know of the situation.

On request we will provide a short training deck and a records template your compliance team can retain as evidence of the briefing. Write to info@yurify.co.

Requests are validated, rate-limited per user, and screened by an output-safety layer before display. We keep operational logs of system events and of privacy-relevant actions, without retaining analysis content longer than the retention periods in our privacy policy. Users can delete their own material at any time from your data and rights.

Workplace communication is personal data of both the user and the people mentioned. Our DPA covers processing terms, and we will supply the information your data protection officer needs for a data protection impact assessment, including processing purposes, categories of data, retention, sub-processors, transfer mechanism, and security measures. Where a works council or Betriebsrat consultation applies, we can provide a plain-language system description for that process.

If Yurify produces output that appears unsafe, discriminatory, or otherwise wrong, report it to info@yurify.co with the date and a description. We acknowledge within two business days, investigate, and where an incident is reportable under Article 73 we notify the relevant authorities and affected deployers. You can also flag any individual result inside the product using the feedback control shown with the output.

The prohibitions in Article 5 and the AI literacy duty in Article 4 have applied since 2 February 2025; transparency and general-purpose model obligations apply from 2 August 2025, with the remaining provisions phasing in through 2026 and 2027. We review this page as guidance and harmonised standards develop, and will update the date above when it changes.

Yurify Pte. Ltd., 168 Robinson Road, #20-01, Capital Tower, Singapore 068912. Compliance and data protection enquiries: info@yurify.co.