Elfbot — Trust, Data Protection & AI Compliance

1How we handle personal data

Public comment moderation, scoped narrowly, with clear legal bases and reviewable retention.

What Elfbot processes

Public comment text, commenter name/handle and profile picture as shown on the platform, page-level metrics, and the actions your reviewers take. Elfbot does not request private messages, friend lists, or off-platform data.

Legal basis

Legitimate interest (Art. 6(1)(f)) for moderating comments already public on your page, balanced against commenters' rights — and your organization's own consent via Meta OAuth for connecting a page in the first place.

Data minimization

Only the account(s) an admin explicitly connects are ingested. Workspaces are isolated — one client's comments, guidelines, and team are never visible to another.

Retention

Comment and moderation-action records are retained for the pilot/contract term to support the audit trail below, then deleted or returned on request. Retention windows are configurable per deployment.

Data subject rights

Commenters (and your own team members) can exercise access, rectification, erasure, restriction, portability, and objection rights. Requests route through the contact below and are actioned within statutory timelines.

Sub-processors & transfers

Comment content is sent to: Meta (source platform), an AI classification provider, and — only when a reviewer requests a fact-check — a web-search provider (Tavily). Transfers outside the EEA rely on Standard Contractual Clauses; a current sub-processor list is available on request.

Breach notification: in the event of a personal data breach, affected organizations are notified without undue delay and, where required, within 72 hours of Elfbot becoming aware — consistent with Art. 33.

EU AI Act

2How the AI is governed

The classifier recommends; it does not decide. Every visibility change traces back to a signed-in reviewer.

Design principle: Elfbot's classifier flags and explains — it never hides, deletes, or replies on its own by default. A human reviewer reads the AI's reasoning and either confirms or overrides it before anything changes on the live page.

Human oversight (Art. 14)

New deployments start in dry-run: the classifier suggests actions but nothing is auto-hidden until an admin deliberately enables live automatic hiding. Reviewers can re-run the classifier, escalate, or fully override any suggestion.

Transparency (Art. 13 / 50)

Every flagged comment shows the classifier's reasoning next to the text, not just a verdict. Reviewers see why a comment was flagged and can request cited, sourced evidence before acting on a factual claim.

Record-keeping (Art. 12)

Every classification, escalation, hide/unhide, and reply is logged with the reviewer, timestamp, and prior AI recommendation, and can be exported as an audit log for accountability reviews.

Accuracy & robustness (Art. 15)

Fact-check claims are answered with cited sources rather than an unqualified verdict, and reviewers are explicitly shown when a claim is a subjective opinion rather than a checkable fact — reducing false-positive moderation of political speech.

Risk classification note: comment moderation on a customer's own page is not among the Annex III high-risk use cases (employment, essential services, law enforcement, etc.). Given the political/civic pilot context, Elfbot is nonetheless built with high-risk-grade safeguards — mandatory human sign-off, full audit trail, and no autonomous public-facing action — by default rather than by exception.

Platform Policy

3Meta / Facebook platform compliance

Access is API-based and admin-authorized; Elfbot never impersonates the page or acts without a human in the loop.

Official Graph API only — pages are connected through Meta's OAuth consent flow with explicitly granted, admin-revocable permission scopes. No credential sharing, no scraping.
No autonomous posting. Replies published to Facebook are always composed or approved by a human reviewer before they go live — Elfbot does not generate and publish public replies unattended.
Moderation actions mirror Meta's own tools. "Hide" and "delete" map to native comment-moderation capabilities available to the page admin; Elfbot does not bypass or duplicate Meta's abuse/report mechanisms — the escalation panel routes serious cases to Facebook's own block/report flows.
Rate-limit respectful. Comment ingestion is event-driven via webhook subscriptions, not high-frequency polling, in line with Meta's platform terms.
Revocable at any time. An admin can disconnect a page from the Your Pages screen, which stops ingestion and access immediately.

4Security & data residency

Encryption	In transit (TLS) and at rest for stored comments, credentials, and audit records.
Isolation	Per-workspace data isolation; team members only see the client workspace(s) they're invited to.
Access control	Role-based team access (owner, member, billing admin, viewer) with Google OAuth sign-in — no shared passwords.
Auditability	Every moderation decision is attributed to a reviewer and exportable as an audit log.
Hosting region	`fill in: EU region / provider`

5Notes for NGO & political pilots

Election-sensitive content: during campaign periods, we recommend keeping automatic hiding off (dry-run) and routing anything ambiguous to a second reviewer — Elfbot's guidelines and escalation tools support this, but the setting is yours to choose per pilot.

Freedom of expression: Elfbot's fact-check tooling is scoped to verifiable factual claims and explicitly flags when a comment is opinion, not a checkable claim — moderation guidelines and thresholds remain under your organization's editorial control, not Elfbot's.

Data protection contact & DPA

A signed Data Processing Agreement (DPA), current sub-processor list, and this summary in PDF form are available on request before a pilot begins.

Data controllerLogiq Media

Data protection contactadmin@logiq.media

EU representativefill in if outside EU

Request a DPAdpa@yourdomain.eu