🛡
Child Safety
CSAM detection
All uploaded images and videos are scanned against globally verified hash databases and image-based classifiers. On match, content is blocked instantly and authorities are notified.
// manifesto
Can an algorithm-free social network use AI? Yes — but where, and why?
Mio's core promise is simple: there's no 'For You' algorithm deciding what you see, exploiting your attention or building addiction. AI does not rank your feed.
But AI is a tool far beyond that. The same technology, instead of exploiting attention, can be used to protect you from harassment, scams, child abuse content and harmful material. We do that — and on this page we transparently document where we use it and where we will never use it.
// In production today
8 AI systems we're using right now
All for safety, accessibility or user benefit. None for feed ranking or ad profiling.
live
⚠
NSFW Filter
Adult content detection
Images, video frames and profile photos pass through a real-time NSFW classifier. Sensitive content is blurred or fully blocked, kept behind an age gate.
🤖
Spam & Fake Accounts
Spam and bot detection
Signup flows, messaging patterns and behavioral signals (device fingerprint, IP pool, signup velocity) flag bot/spam accounts — auto-suspended and routed to human moderators.
💬
Text Moderation
Hate speech & harassment detection
Comments, messages and posts are classified in multiple languages for hate speech, harassment, threats and sexual content. Above-threshold content is escalated to human review — no auto-deletion.
🌐
Translation
Automatic content translation
Comments and posts can be translated to the user's language on demand. Translation is opt-in and does not modify the user's original text.
🎙
Accessibility
Voice message transcription
Voice messages can be transcribed to text for hearing-impaired users on demand. Data is processed per-message and not stored.
🔍
Visual Understanding
Auto tag suggestions
Uploaded photos/videos suggest topics, locations and themes — but tags are never added unless the user confirms. The user decides, not the algorithm.
🚨
Crisis Response
Suicide/self-harm signal detection
Messages or posts containing risk signals trigger a private, non-judgmental support message with local crisis hotline information.
// what we don't do
Where we will NEVER use AI
🚫
NO AI in feed ranking
There is no 'For You' algorithm on the main feed. Order is driven by tag match, interests and chronology. You decide what you see — not us.
🚫
NO engagement maximization
We do not run AI systems engineered to extend session time via rage bait, polarizing content, or addiction loops — and we never will.
🚫
NO AI ad profiling
Mio is ad-free. We do not train or use any AI to profile users for ad targeting purposes.
🚫
We DO NOT sell user data to third-party AI training
User content, messages, images or behavior is never sold, shared or licensed to third-party AI companies as training data.
🚫
NO automated content deletion
No user content is deleted by AI alone. AI generates signals; the final decision always passes through a human moderator. The user always has the right to appeal.
// coming soon — 2026 Q3 / Q4
AI features shipping in the next 6–9 months
Each solves a real user problem. None of them shipped just because 'AI is trendy'.
🌍
2026 Q3Moderation in 50+ languages
Expand hate speech and harassment detection to 50+ languages. Contribute to open-source models in low-resource languages (Turkish dialects, Arabic dialects, Kurdish variants).
🎭
2026 Q3Deepfake & AI-generated content labeling
Automatically detect and visibly label AI-generated images and videos. Mandatory transparency label — cannot be hidden.
✍
2026 Q4AI Writing Assistant (opt-in)
Optional AI assistant that improves drafts, corrects grammar and suggests translations on demand. Off by default. No post is ever auto-generated — all generation is explicit with user consent.
♿
2026 Q4Auto alt-text (50+ languages)
Auto-generated alt-text on all images for vision-impaired users. The creator can always edit the alt-text.
📞
2026 Q4Real-time translation in voice calls
Real-time multi-language translation in voice messages and live broadcasts. Data is processed per session and never recorded.
// long-term — 2027 and beyond
From Türkiye to the world — our responsible AI vision
We're aiming beyond 2 years too. Open-sourcing our child safety models, keeping data on-device with federated learning, contributing to local LLM ecosystems — these aren't 'maybes', they're our roadmap.
🛡
Mio Safety AI — open-source child safety model
We plan to open-source Mio's child safety classifiers and training methodology — a Türkiye-to-world contribution so that small platforms can be just as safe as the giants.
🔐
Federated Learning — keep data on device
Train models directly on user devices without ever sending raw data to our servers (federated learning). We want to prove safety doesn't have to come at the cost of privacy.
📊
Explainable AI (XAI) — a 'why' for every decision
If your content was hidden/warned/removed, an Explainable AI layer answers 'why' — in human language, not technical jargon.
🇹🇷
Türkçe-focused local LLM partnerships
Financial and data contributions to open-source Turkish-focused LLMs (TURNA, BERTurk, etc.) — strengthening the independent Turkish AI ecosystem.
🌐
Mio Trust Graph — anonymous social trust score
A community-verified trust layer based on social interaction patterns — without using personal identity. Structural defense against fake accounts, scams and manipulation.
🤝
User-portable AI assistant
A personal AI assistant scoped only to your own content, fully under your control, portable to other platforms. Disable-able, delete-able, exportable.
🧠
Algorithmic independence audit (3rd party)
Annual third-party audits of our AI systems with publicly published reports. Transparency proven by an independent body — not just by our word.
🎓
Mio AI Ethics Council
An independent council of academics, rights advocates and user representatives reviewing major AI features before launch.
// Mio AI principles
The 6 principles every AI feature must pass
These principles are part of our product process. If an AI feature can't meet even one, it doesn't ship.
01
Transparency is the default
We publicly document which AI we use, where, what decisions it makes and what its limits are. This page is proof of that promise.
02
User control always wins
Users can appeal every AI decision. There is no auto-deletion; a human moderator makes the final call.
03
Data minimization
AI only processes the minimum data needed for its task. Raw content is discarded after classification — only metadata is logged.
04
Generation vs. detection separation
We use AI to make content safe, NOT to generate content. AI generation features are always opt-in and clearly labeled.
05
Open source & auditability
We prefer open-source models when possible. When using closed source, we publicly disclose the provider and data policy.
06
No manipulation, only service
We will NEVER use AI to make users addicted or to manipulate them. AI exists to serve the user — not to extract data from them.
// for researchers & media
Need more detail?
For details on our AI policies, models used, audit reports and data-handling protocols, please reach out. Academic research requests are prioritized.
// user pledge
AI is a tool for us. You are not a customer — you are the user.
We don't sell data. We don't show ads. Our AI exists to protect you — not to monetize you.
// AI FAQ
Frequently asked questions
// why this page
Transparency is proof, not a claim
Most social networks say "we care about AI ethics" but never document which model, for what purpose, within which limits. This page is that documentation.
Mio's AI approach will go through annual independent audits; reports will be publicly published. By 2027, we aim to open-source the core architecture of all our safety models.
For questions, feedback and academic collaborations: [email protected]