AI Use Disclosure
Last updated: 2026-05-23
Constitution §1.5 requires SocialAI to be transparent about which AI providers process your data and how. This page lists each provider and what they see. Replace this draft with your final legal copy, reviewed by counsel, before going live.
1. What AI does on SocialAI
- Caption + content generation — Anthropic Claude (Haiku, Sonnet) and OpenAI GPT-4o / 4o-mini.
- Image + video generation — Black Forest Labs FLUX (via Replicate) and OpenAI gpt-image-1.
- DM and comment replies — Anthropic Claude via our self-hosted Python brain service.
- Vector memory + knowledge-base search — Voyage AI embeddings stored in our own ChromaDB instance.
- Analytics insights — Anthropic Claude generates narrative summaries of your analytics dashboard.
2. What data is sent to which provider
| Provider | Receives | Zero-retention |
|---|---|---|
| Anthropic | Your caption prompt, brand kit voice samples, knowledge-base excerpts, the user's DM (if running an agent). | Yes — enabled on our enterprise contract. |
| OpenAI | Your caption prompt and (optionally) image-generation prompts. | Yes — enabled via API opt-out flag on all our requests. |
| Replicate | Image-generation prompts only. | Yes — we delete the generated assets from Replicate's storage after download. |
| Voyage AI | The text of knowledge-base documents you upload, in chunks, for embedding. | Yes — embeddings only, no retention of source text. |
3. What we do NOT send to AI providers
- Personally identifying information (PII) — emails, phone numbers, names — unless you explicitly include them in your prompts.
- OAuth tokens, payment methods, or other secrets — those never leave our infrastructure.
- Other workspaces' data — every AI request is scoped to your workspace.
4. Bot disclosure to your audience
When SocialAI replies to a DM or comment on your behalf, it does so as your account. You are responsible for disclosing to your audience that automated replies are in use, as required by platform policies (Meta's Bot Policy, etc.) and any local law (EU AI Act, etc.). See our Acceptable Use Policy.
5. Training
SocialAI does not train its own foundation models on your content. Our AI providers are configured to not use your data for training wherever that opt-out is available (Anthropic enterprise, OpenAI API zero-retention).
6. Your controls
- You can turn off AI features per workspace in Settings → AI.
- You can disconnect any social account at any time.
- You can request export or deletion of all your data via Settings → Account → Data export / delete.
7. Changes
We update this disclosure whenever we add or remove an AI provider. Material changes are notified 30 days in advance via email and in-app banner.