EU AI Act
Last updated: 29 September 2026
This notice explains how MeiMei Phone Assistant meets the transparency obligations of the EU AI Act (Regulation (EU) 2024/1689), in particular Article 50, which applies from 2 August 2026.
MeiMei Phone Assistant is an AI system that interacts directly with natural persons. On a phone call it transcribes the speech of the caller, generates answers with a language model and speaks them back; in the optional realtime voice mode (preview, off by default) a voice model of the provider chosen by the operator listens and answers directly. It can also greet callers and take messages.
The system is not a prohibited practice and is not classified as high-risk: it does not fall under any Annex III use case such as employment, credit scoring, or emergency triage. It is an AI system that interacts with natural persons, so the Article 50(1) transparency duty applies: people must be informed that they are interacting with AI, unless this is obvious.
The assistant can identify itself as an AI assistant at the start of the calls it handles, for example: "You are speaking with an AI assistant, not a human." The disclosure is off by default: the operator turns it on with the setting PHONE_ASSISTANT_AI_DISCLOSURE=on.
With the disclosure on, the built-in sentence follows the device language (English, Romanian, French, German, Spanish or Chinese; otherwise English) and precedes the configured greeting or outgoing-call script; in the realtime voice mode the voice model is instructed to say it first. Operators can change its wording or its position relative to the greeting.
In MeiMei Cloud Calling the AI disclosure is on by default: at the start of every call the assistant handles, callers hear in the device language that they are speaking with an AI. We recommend keeping it on; you can change its wording.
Full-duplex voice models such as GPT-Live-1 listen and speak at the same time, so callers can interrupt the assistant at any moment. The disclosure is spoken before the conversation starts, and transcripts of cloud calls are kept only in your MeiMei account.
If you choose MeiMei AI, the answers you hear come from AI models that we enable and run under our own accounts with Google (Gemini models) and OpenAI. They can be speech recognition, language and speech models, or a live voice model. The models on offer are listed in your account and can change.
Nothing changes for callers: the AI disclosure works as described above (on by default in MeiMei Cloud, off by default on an appliance), and callers can always be told they are speaking with an AI. With MeiMei AI we run more of the service ourselves, so we act as the provider of the service; the obligations of the providers of the general-purpose models stay with Google and OpenAI. Your own-key setups are unchanged.
If you operate the device and the EU AI Act applies to you (Article 50(1), applicable since 2 August 2026), you must inform callers that they are speaking with an AI unless this is obvious: turn the AI disclosure on, or inform callers in another way. After changing greeting or disclosure settings, make a test call and check that callers hear the disclosure; and check the national rules on call recording and consent that apply in addition to the AI Act.
Article 50(1) does not require disclosure where it is obvious to a reasonably well-informed person that they are interacting with AI. Do not rely on this exception unless it clearly applies to your setup.
Article 50(2) requires machine-readable marking of synthetic content such as deepfakes. The assistant produces conversational speech for a live call and does not generate or manipulate image, audio, or video content intended to be passed off as authentic, so the deepfake rules do not apply.
The assistant uses third-party general-purpose AI models (GPAI) chosen by the operator, for example hosted LLMs or local models. The GPAI provider obligations that apply since August 2025 rest with those model providers.
Questions about AI compliance: support@meimeiassistant.com.