กลับไปหน้าภาพรวม

EU AI Act (AI Regulation)

อัปเดตล่าสุด: 17.11.2026 เวอร์ชัน 2.0 TH

EU AI Act

The EU AI Act (Regulation (EU) 2024/1689) has been in force since 1 August 2024 and applies in stages, fully from 2 August 2026. This page separates what runs on our own rules from which AI models are actually used. The individual features are described on the page AI at Friendstime.

1. Our own rules and filters (local, no AI models)

Word filters and abuse and spam protection work by fixed rules on our own servers. They are not trained AI models and send nothing to third parties. Account suspensions and hiding content are decided by people, not by an automatic system or an AI.

We also read image provenance data (IPTC “Digital Source Type”, Content Credentials/C2PA, markers of known image generators) locally from the file. If something is found, or you mark the image yourself, the image shows “AI-generated” or “Edited with AI”.

2. AI models that actually run

FriendsBot (chat): the replies are written by an AI model that runs on computers of the operator, not at Google or another large provider. Only your messages in the FriendsBot chat are sent there.

Google Gemini (Google, USA): the connection is switched off in the code. No data is currently sent to Gemini. AI translation, reply, hashtag and caption suggestions, the automatic AI fact check and the AI image check are therefore not running at present.

There is no connection to OpenAI.

3. Order in the feed

The default is “Recommended”. The starting point is the time of the post; posts with more interactions move forward by up to eight hours. There is no interest profile built from your behaviour for this. A purely chronological order with the newest first is chosen on the web with “Newest” or in the settings under “Appearance”, and in the app with “Chronological feed”. More: How your feed is put together.

4. What we do not do

No real-time biometric identification and no face recognition. No social scoring. No analysis of your mood when you write. No predictions about criminal acts of individual users.

5. Training and data

We do not train AI models with your posts, photos or messages. We would only do so with your prior consent. The word filters and rules under 1. are not trained models; they have no training data.

Human review and complaints

If a post was removed or an account restricted, you can ask for a review by people: Reporting, moderation and appeals. You can send complaints about AI use to us (ai@friendsti.me) or to the competent authority, in Germany the Bundesnetzagentur.