Security & AI

How We Protect Your Data

• Row-Level Security (RLS): Every database table has access policies that ensure users can only read and modify data they're authorized to access.
• Secure authentication: Sign-in uses Google or a one-time email code — we never see or store a password. The sign-in form does not reveal whether an account exists.
• Closed boards: Personal and team boards are visible only to approved members, and their images are served through short-lived signed links.
• Moderation: Every published or edited wish is checked on the server against a wordlist and an AI classifier. Two reports from different users hide a wish until it is reviewed. Rate limits bound publishing, editing and AI use.
• Protected API keys: Only a public (anon) key is exposed to the browser. All sensitive operations happen server-side in secure backend functions.
• Security Definer functions: Critical operations like updating spark counts use secure database functions that bypass direct table access, preventing unauthorized modifications.

Editing a wish

The owner can edit their wish. Once other learners have joined it, only a small correction is accepted; a bigger change should be published as a new wish. Edits go through the same content check, earlier versions stay visible, and joined learners are notified. The editor of an anonymous wish is never revealed.

Administrator rights

• Administrators can edit public wishes with the same limits as owners, delete wishes and regenerate images.
• Administrators do not have these rights on closed team boards or personal boards.
• Administrator deletions are recorded in an audit log.
• Roles are stored separately from user profiles and checked on the server.

What We Don't Collect

• No marketing cookies or third-party trackers
• No IP address logging for profiling
• No behavioral analytics or heatmaps
• No data sold to third parties — ever

Infrastructure & Data Location

The service's database and file storage are located in the EU (Ireland). For us, digital sovereignty means following European legislation and values and keeping our freedom of choice: we avoid vendor lock-in and choose portable, replaceable solutions. Where possible, we prefer European systems.

Security Auditing

We maintain an internal security document (SECURITY.md) that tracks identified risks, their severity, resolution status and consciously accepted risks. Findings are reviewed and fixed continuously.

Our Principles

• Human decides, AI assists — AI generates inspiration and images, but you always choose what to post, share, or delete.
• No profiling, no scoring — we do not evaluate, rank, or classify users in any way.
• Your data doesn't train AI models — your wishes and personal information are never used to train or fine-tune AI models.
• Transparency — AI-generated content is always clearly indicated (images, inspiration stories). You always know when AI is involved.

EU AI Act — Risk Classification

Under the EU AI Act, AI systems must be classified by risk level. Spark's Wish Wall is classified as a Minimal / Low-risk AI system.

Why:

• The system does not make decisions on behalf of people
• The system does not affect users' rights, obligations, or legal status
• The system does not use personal data for evaluation or prediction
• No automated profiling (GDPR Art. 4(4))
• No automated decision-making that affects people

What Spark Does — and What It Doesn't

Prototype Stage — Honest Transparency

Spark's Wish Wall is currently a prototype. We comply with EU legislation and European values in how we handle data and AI.

For the full vision of our approach to trustworthy AI, see SwarmMind AI Safety & Trust.

Roadmap

• Ongoing review of European infrastructure and hosting options
• Formal EU AI Act risk assessment document (PDF)
• User data export (GDPR portability)
• Content moderation transparency report
• Accessibility audit (WCAG 2.1 AA)