Where we use AI — and where we don't.

1. Use cases — where AI is deployed

OGRECORDS™ uses AI in three clearly scoped technical domains today:

• Audio QC: detection of clipping, phase issues, loudness (LUFS / true-peak), silence, sample-rate and bit-depth compliance.
• Cover QC: pixel resolution, file format, likelihood of trademark or celebrity recognition as a hint for human review.
• Genre and mood tagging: suggestions based on audio features that the artist can accept or reject.

2. Where AI is deliberately NOT used

We never use automated decision-making for the following — these are reviewed manually by the operator:

• Calculation or allocation of royalties and splits.
• Approval or denial of payouts (manual SEPA / PayPal workflow).
• Content moderation, account suspensions or strikes without human review.
• Classification of sync requests or pricing decisions in the sync marketplace.

3. Models and providers

We rely exclusively on models from established providers with explicit privacy and retention guarantees. Today we use AWS Bedrock models (Anthropic Claude, Amazon Titan) for inference. Calls go through our server-side AI gateway, never directly from the user's browser.

No calls go to providers without a data processing agreement (DPA) in place.

4. Training data and user content

We do not train any of our own models on uploaded audio, cover art, metadata or personal data of our users. User content is processed only to deliver distribution (storage, DSP delivery, smart links).

Our inference providers also do not train on content sent through our pipeline under their DPA terms.

5. Transparency, correction and opt-out

AI suggestions (e.g. genre tag, mood) are explicitly labeled as such in the UI. You can override them at any time. AI-based audio or cover warnings never auto-reject a release — they only open a review ticket for the operator.

If you want to opt out of having your content flow through our inference pipeline, email info@og-records.com — we'll provide a manual review without the AI step.

6. Security, logs and retention

All AI calls are logged server-side in a pseudonymized form (track ID, model, latency, token count) — never the content itself. Logs are deleted automatically after 90 days.

We do not run on-device or in-browser AI for sensitive data.

7. AI chat bot on the website

We run an AI chat bot on og-records.com (and our Telegram + WhatsApp bots) that answers questions about the platform. The bot identifies itself as an AI before every new conversation.

• The bot is not legal advice, not tax advice, not a binding statement. Answers can be wrong.
• Binding statements only come via info@og-records.com.
• The bot uses a Retrieval-Augmented-Generation (RAG) pipeline over the public website content and our legal markdown documents. It has no access to user accounts, royalty data or payouts.
• Chat content is stored temporarily for quality assurance and deleted after 30 days. Inputs are NOT used to train any model.

8. Marketing copy and cover suggestions

Some marketing copy (press blurbs, smart-link subtitles, release notes) is partially generated with AI assistance and then reviewed editorially. We do NOT use AI-generated marketing images or covers as the default on the platform.

If you explicitly upload an AI-generated cover, that's your decision and falls under our distribution terms (master rights, DSP rules). We check covers only for technical compliance, not for AI provenance.

9. No fully automated decisions (GDPR Art. 22)

We do not take decisions producing legal effects against you based on automated processing alone. Specifically: AI models do not decide about payouts, advance approvals, terminations, strikes or DSP takedowns. Those decisions are always taken by the operator manually — AI signals at most flag something for a human review.

10. Sub-processors and data residency

Inference calls run exclusively through AWS Bedrock (region eu-central-1, Frankfurt) and do not go to OpenAI, Google Gemini, xAI or comparable providers without an EU DPA.

• AWS Bedrock — model inference (Anthropic Claude, Amazon Titan, Amazon Nova).
• AWS S3 (eu-central-1) — audio and cover files for the QC pipelines.
• Cloudflare Turnstile — bot protection for the chat widget (cookie-free).

The current list of sub-processors with contracts and data flow is in the privacy policy.

11. Limits and risks

AI models can be wrong. Concretely:

• Audio-QC findings are statistical — a track marked as "clean" can still have mastering problems.
• Cover-QC does not detect every trademark conflict; final responsibility lies with the artist.
• Genre tags are suggestions, not binding classification.
• Chat-bot answers can be out of date, incomplete or simply wrong. Binding statements only by email.

We improve our pipelines continuously and welcome feedback at info@og-records.com.

12. Changes

This policy describes the current state of our AI usage. We update it on a rolling basis as new use cases come up or existing ones change — changes apply from the moment they are published on this page, without separate notice.

Questions about how we use AI?

If you want to know how a specific use case is implemented, or you need an audit trail (e.g. for a label or compliance request), email us — we reply manually.

info@og-records.com