Privacy Policy#

This page is a plain-language template drafted for RTFC's pre-launch, invite-only phase. It has not been reviewed by outside counsel; a signed Data Processing Agreement always takes precedence over this page for a specific customer.

Last updated: October 2026.

Controller and processor#

Faktolumo SAS is the data controller for account, billing and usage data (see Costs). Where a customer submits its own end users' audio or transcripts for fact-checking, that customer is the controller of that content and Faktolumo acts as processor on its behalf, under the Data Processing Agreement.

This page describes both: what we process about you as a direct account holder, and how we handle content a customer asks us to process on its behalf.

What we process#

Depending on how RTFC is used, we process:

• Account data - username, email, role, API tokens, plan and subscription, billing address (via Stripe).
• Submitted content - audio recordings, transcripts, detected claims and their context, research results and cited sources.
• Usage and cost data - which endpoints were called, token/search/transcription volumes, and the resulting billed amount (see Costs).

Voice biometrics#

When diarization or voice identification is enabled for a debate, RTFC extracts short voice embeddings - mathematical representations of a speaker's voice, not a recording - to tell speakers apart within a session and, where enrolled, to match a voice to a named person. This is special-category biometric data under Article 9 GDPR.

The people whose voices are processed this way are, in practice, participants in a recorded broadcast or debate rather than Faktolumo's own contracting customer, so the applicable legal basis is not their individual consent but our legitimate interest - and that of the customer publishing the fact-check - in accurately attributing a public statement to the person who made it (GDPR Article 6(1)(f), read together with the journalism and freedom-of-expression accommodation in Article 85). We do not use voice biometrics for any purpose beyond speaker attribution within the debate being fact-checked.

Voice embeddings and the underlying audio slices are retained for 7 days from creation - the shortest window that still lets an operator correct a misattributed speaker before the material is deleted. See Data retention.

Brand checks: public content of creators#

When you run a brand check, a monitor or an account view, RTFC processes publicly available content of third parties who are not users of the service: a creator's public handle and public profile details (name, follower count, verification badge), the audio of their public videos (transcribed by a speech-to-text provider), the captions, hashtags and accounts they cite, and the passages that mention a brand, with a tone and a stance assigned by an AI model.

For this processing the customer who runs the check decides the purpose and is the controller; Faktolumo acts as its processor (see the Data Processing Agreement). To avoid transcribing the same public video twice, Faktolumo keeps the transcript and the scan of a public video on its own account and may reuse them for another customer who analyses the same video; the passages shown to a customer are those of the check that customer ran.

No voice identification is run and no voiceprint is created or matched for brand-check videos. Private accounts and content behind a login are not accessed.

A creator whose content was analysed can ask to see it, to have it erased or to object by writing to contact@faktolumo.eu. We then delete the stored transcripts, scan results and account record of that person.

Data retention#

Retention periods are table-specific and tied to why the data is kept, not one blanket window: operational and audit logs (365 days), the transcript/claim archive (5 years by default, reflecting its role as a factual record), and voice biometrics (7 days, see above).

A specific plan or a signed agreement may set a different retention window for some data than the defaults above - for example a shorter window on request, or a longer one where a customer's own contract requires it - never below the floor a special or sensitive data category requires.

Your rights#

Subject to French and EU data protection law, you may request access to, rectification of, or erasure of your personal data, object to or restrict its processing, and request portability of data you provided directly.

How erasure works in practice. Erasing an account never destroys the billing records tax law requires us to keep, for up to ten years in France. What we erase is content - transcript and claim text, voice biometrics, and the free-text portion of AI and search logs. What remains is the numeric record of what was billed, when, and to which now-anonymized account - a record that carries no content of its own, kept under the legal-obligation basis in Article 6(1)(c) GDPR (Article 17(3)(b) exempts it from the erasure right). The invoice itself, with any billing name and address tax law requires, is retained by our payment processor, Stripe, under its own compliance program, not in RTFC's own database.

To exercise any of these rights, contact contact@faktolumo.eu. Requests are handled by an administrator; we may ask for information to verify your identity before acting on a request.

Where data is processed#

RTFC's own infrastructure runs on EU-based data centers. A number of the AI, search and speech-to-text providers we use to produce a fact-check are themselves located outside the EU/EEA - see Subprocessors for the full list and, where publicly known, each provider's location.

Where a subprocessor is located outside the EEA, the transfer relies on that provider's own Standard Contractual Clauses or another GDPR-recognized safeguard under its own data processing terms. We have not independently audited every subprocessor's transfer mechanism; a customer with a specific requirement here should contact us before relying on this page for a compliance decision that depends on it.

How AI providers handle your content#

The step that extracts claims directly from your transcript is restricted to AI providers that contractually commit not to train their models on submitted data. This restriction currently applies to that one step; a later step that researches and drafts the verdict for an already-extracted claim is not yet restricted the same way, and may use a wider pool of providers. See Subprocessors for which providers are in the no-training pool.

This is a technical control we can verify and improve over time, not a blanket guarantee that no AI provider ever sees your content - if that distinction matters for your use case, contact us before relying on it.

Cookies#

The RTFC dashboard sets one cookie: a session cookie used to keep you signed in. It is strictly necessary for the service to function and is not used for analytics, advertising or tracking, so it does not require your consent under the ePrivacy rules - only disclosure, which is what this section is.

We do not currently run any analytics or tracking script on the dashboard or on this documentation site. If that changes, this page - and a consent banner where the new use requires one - will be updated first.

Changes to this policy#

We may update this policy from time to time; material changes will be reflected here with an updated date.

Contact#

Privacy questions and data-subject requests: contact@faktolumo.eu.