Trust & Safety at Velvet.AI

Last updated with the legal documents · Effective date: July 8, 2026

Velvet.AI is an adults-only (18+) platform for chatting and roleplaying with fictional AI characters. Adult content between consenting adults and clearly-adult fictional characters is permitted; content that exploits or endangers minors, depicts real people without consent, or is otherwise illegal is not — ever. This page explains, concretely, how we keep it that way.

Who operates this service

Velvet.AI is operated by Emil Svensson (sole operator). We respond to safety, legal, and law-enforcement inquiries at legal@velvetapp.ai, and to privacy matters at privacy@velvetapp.ai. Our full Terms of Service, Privacy Policy, and Content & Conduct Policy are public and require no account to read.

Adults only

• Every visitor must pass an age gate — an explicit 18+ confirmation combined with an automated anti-bot check — before reaching the service.
• The service is not offered in territories whose law requires age-verification measures we have not yet implemented (currently the United States, the United Kingdom, Germany, France, Italy, and Australia — enforced by geolocation at the network edge).
• We do not knowingly permit anyone under 18 to use the service; accounts we learn or reasonably suspect to be under-age are closed and their data deleted.
• We are committed to adopting risk-appropriate age-assurance measures as the service grows, and plan to adopt EU-blueprint / EUDI-wallet age verification as it becomes available.

Zero tolerance for child sexual abuse material (CSAM)

Any sexual, sexualized, or suggestive content involving a minor — or a character that is minor-coded or childlike — is absolutely prohibited, with no fiction, "aged-up label", satire, or artistic-intent exception. A character's age is judged by overall impression, not by a stated number, and the prohibition applies regardless of whether such material is illegal in any particular jurisdiction. Confirmed violations result in immediate removal and quarantine of the content, a permanent ban, and a report to the Swedish Police Authority (Polisen) — the primary reporting channel for this Swedish-operated service — and additionally, where there is a connection to the United States or where it is necessary to protect an at-risk child, to the National Center for Missing & Exploited Children (NCMEC), to which we report voluntarily as a non-US provider.

How moderation actually works

Our pipeline has five layers:

1. Automated minor-safety screening of all text. A dedicated automated filter scans character fields at creation, comments, and every chat message before it is sent. It looks for the co-occurrence of minor-indicative and sexual signals, including normalising attempts to disguise words. The check runs in the client for immediate blocking and is re-run authoritatively on our servers, so it cannot be bypassed by a modified client. Blocked attempts are logged for moderation review.

2. Automated image moderation before publication. Every image attached to a public character is scanned server-side with industry image-safety classification (Google Cloud Vision SafeSearch) before it can be published. The gate fails closed in production: if the scan cannot run, the content is not published. Free-form user image upload is currently disabled; catalog imagery is curated. We are additionally onboarding industry hash-matching tooling (PhotoDNA / NCMEC hash lists) to strengthen detection of known CSAM.

3. Human review of published characters. New public characters enter a pending-review state. Nothing leaves that state on age alone: automatic release happens only after passing a fresh automated content re-scan, and a hard safety failure is deleted and logged instead. Moderators can approve, reject, or remove content and act on accounts at any time.

4. User reporting. Any signed-in user can report a character or comment in-app. Reports require a category (including minor-safety, non-consensual or illegal sexual content, harassment, and spam) and a written explanation, and are triaged by our moderation team.

5. Enforcement and escalation. Depending on severity and history: content removal, feature restriction, temporary suspension, or permanent ban. For CSAM we operate a documented internal escalation runbook: detected material is immediately quarantined in an encrypted, access-restricted evidence store and purged from all serving layers; a single designated reviewer makes a one-time, logged determination; confirmed cases are reported to Polisen within 24 hours (and to NCMEC where there is a US connection or a child at risk); evidence is preserved for 90 days after a report (12 months for NCMEC reports, or longer under a law-enforcement preservation order) and then irretrievably deleted, keeping only a cryptographic hash to block re-upload. Where we remove content or restrict an account, we give a statement of reasons where reasonably possible and offer an internal appeal route with human review.

Other protections

• Real people: non-consensual sexual or intimate content depicting real, identifiable people (including deepfakes) is prohibited and removed.
• Anti-abuse controls: anti-bot verification at the gate, per-user rate limiting, and automated spam/abuse detection.
• Data protection: how we handle personal data — including moderation records — is set out in our Privacy Policy.

Contact

• Safety concerns, law-enforcement and legal inquiries: legal@velvetapp.ai
• Privacy requests: privacy@velvetapp.ai
• Copyright/IP notices: copyright@velvetapp.ai

If you believe a minor is using the service or that content on the service endangers a minor, contact legal@velvetapp.ai immediately.