> ## Knowledge Base Index
> Fetch the complete knowledge base index at: https://knowledge.autocut.com/sitemap.xml
> Use this file to discover available pages before exploring further.
> Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash).

# Does AutoCut or its providers use my data to train AI models?

AutoCut and all of our subprocessors **do not use** your audio, video, transcripts, captions, or any customer content to train AI/ML models. Your data is used **only** to deliver the features you request, then it’s deleted from our cloud after processing.

# Our commitment
* **No training:** We never use customer content (audio, video, transcripts, captions, previews) to train or fine-tune AI models.
* **No human review by default:** We don’t manually review your content unless **you** explicitly share samples for support.
* **Contractually enforced:** All vendors/subprocessors we rely on are **contractually prohibited** from using customer data for model training or unrelated purposes.
* **Local-first design:** Many Voice/AI operations run on your machine. When cloud processing is needed, files are **encrypted**, processed, and **removed** after completion.

# What data this covers (scope)
* **Customer content:** Audio/video you process with AutoCut, plus generated outputs (e.g., transcripts, captions, analysis markers).
* **Operational metadata:** Minimal technical info (job IDs, duration, error codes) needed to run the service and provide support.
* **Account & billing:** Account details (e.g., email, license), with payments handled by Stripe.

None of the above is used for model training.

# How processing works
* **Local vs. cloud:** AutoCut is **local-first**. Certain features may require temporary, encrypted cloud processing.
* **Encryption:** Data in transit uses **TLS**; data at rest (when present briefly in cloud) is stored with **industry-standard encryption**.
* **Retention:** Temporary processing artifacts are **deleted after the job completes**. We keep only minimal, necessary metadata for reliability, billing, abuse prevention, and support.

# Subprocessors & vendors
* We use reputable providers (e.g., cloud hosting/CDN, error reporting, payments).
* Our Data Processing Addenda (DPAs) and vendor agreements **forbid model training** on customer data and restrict processing to **your instructions**.
* Need a vendor list or DPA? Contact us (see below).

# Access & controls
* **Least-privilege:** Employee access is limited and role-based.
* **No human labeling:** We don’t send your files to third parties for labeling or evaluation.
* **Support samples:** If you open a support ticket, you can choose to share example files. We’ll use them **only** to resolve your issue, then remove them.

# How we improve without your data
We rely on **unit tests, synthetic data, public benchmarks,** and **opt-in** test datasets—not on customer content—to improve model quality and features.

# Enterprise notes
* **Firewall rules:** If your company restricts egress, see **“**[Which domains should be whitelisted for AutoCut?](https://knowledge.autocut.com/en/article/which-domains-should-be-whitelisted-for-autocut-1xc1elm/)**”**
* **Privacy reviews:** We can share our security notes, DPA, and subprocessors under NDA for vendor assessment.

# Quick-copy compliance claus**e**

AutoCut and its subprocessors do not use customer content (including audio, video, transcripts, captions, or derived artifacts) to train, fine-tune, or improve AI/ML models. Processing is limited to providing the contracted services, with encrypted transit/storage and deletion of temporary artifacts after completion.