Trust: data, AI and the law
The questions a European business asks before connecting its brands: where the data lives, who else handles it, how AI-made content is marked and who stays in control. Everything here describes how the product works today. Where something does not exist yet, we say so.
Where your data lives and who processes it
The application and its data run on Hetzner servers in Nuremberg, Germany. To write copy, generate images and video and publish, parts of your data go to the providers below. A provider is called when the feature that needs it runs, and a social network is involved only after you connect it.
| Provider | What it is used for | Where |
|---|---|---|
| Hetzner Online GmbH | Servers that run the application and hold its data | Nuremberg, Germany |
| Cloudflare (R2) | Storage for generated images and video, and a separate closed bucket for backups | EU |
| Language models for copy, image generation, speech synthesis and sign-in with Google | USA | |
| BytePlus | Video generation | Singapore |
| fal.ai | Image and video generation | USA |
| TypeSafe AI, Inc. | Structured checks on text, for example whether a post states facts the brand never gave, and sorting incoming comments and messages | USA |
| Apify | Collecting public web data about competitors and audience | EU (Czechia) |
| Stripe | Card payments for plans; checkout pages are hosted by Stripe | See the sub-processors page |
| Mail provider | Service mail only: address confirmation, password reset, invitations | See the sub-processors page |
| Meta, LinkedIn, TikTok | Publishing to your own accounts through their official APIs, once you connect them | See the sub-processors page |
The sub-processors page says what each provider sees. The processing terms are on the data processing page.
How AI-made content is labelled
Article 50 of the EU AI Act, on transparency for AI-generated content, applies from 2 August 2026. This is what SocialOS does today.
- Machine-readable marker. Every image and video SocialOS generates is marked in its file metadata. The IPTC/XMP field digitalSourceType is set to trainedAlgorithmicMedia, with a plain-text note that the file is AI-generated with SocialOS. Video is marked without being re-encoded.
- Your own files are left alone. Files you upload yourself are not marked.
- AI flag on each post. A post has a switch. When it is on, a closing line saying the post was made with AI is added to the caption at publishing, in English, Ukrainian or Polish, chosen from the caption's language.
- Optional visible tag on images. A visible AI tag on text-hook images is a brand setting. It is off by default.
A metadata label is not a C2PA signature, so it does not prove origin cryptographically. SocialOS does not offer C2PA signing yet. Whether a post needs a visible label depends on what it shows. The terms say who is responsible for that choice.
People stay in control
- Approval is the default. With autopilot off, a post due for publishing waits in the approval queue until a person on your team approves it.
- Autopilot is opt-in. It is a separate switch for each brand and stays off until you turn it on. You can switch it off again at any moment.
- A daily cap applies even with autopilot on. The default is 10 posts a day for a brand. Posts over the cap go back to the approval queue. You can also set quiet hours in which nothing is published on its own.
- AI apps publish only if you allow it. When you connect ChatGPT, Claude or another AI app, you choose per connection: publish immediately when you ask it (your confirmation in its chat is the approval) or ask you first. Each connection that publishes immediately is capped at 10 posts and 30 replies a day by default, every action is logged with the app's name, replies carry the AI label (on by default), and you can switch it back or disconnect it at any time.
Your data rights
- Export. A signed-in user can download everything the organisation holds as one ZIP of JSON files. Media files are not inside it, only an index with links. Social-account tokens and password hashes are left out on purpose.
- Delete your account yourself. You confirm by typing a phrase, and the screen first shows what will go. If other people still use your organisation, only your access is removed. If you are the last member, the whole organisation is deleted: brands, posts, media, connected accounts and credits.
- Connected accounts. Before the files are erased, we ask Meta to revoke SocialOS's access to your Instagram and Facebook accounts. LinkedIn tokens are deleted on our side.
- What deletion does not reach. Copies in our backup storage and finished media already uploaded to Cloudflare R2. If you want those removed too, write to us.
How we use personal data and the rights the law gives you are in the privacy policy.
Security basics
- Encrypted tokens. Access tokens for your social accounts are stored encrypted on disk. The key is held in the server's environment, not next to the data.
- Official APIs only. Publishing goes through Meta's Graph API, LinkedIn's API and TikTok's Content Posting API. You connect an account on the network's own sign-in screen, and SocialOS never asks for your social-media password.
- Brand isolation. Data is kept apart for each organisation and, inside it, for each brand. A signed-in user reaches only the organisations and brands they were given access to.
- Passwords and backups. Passwords are stored only as salted PBKDF2-SHA256 hashes. Backups of system state go to a separate, closed bucket, not the one media is served from.
More detail, including sign-in and sessions, is on the security and data page.
Contact
SocialOS is run by Modus Altos Mikhail Kushnir in Kraków, Poland. For questions about data, AI labels or the processing terms, use the contact form or write to mikhail@agenticos.agency.
Privacy · Terms · Sub-processors · Data processing · Security & data
A question we have not answered?
Ask before you connect anything. A person replies, usually the same day.