> ## Documentation Index
> Fetch the complete documentation index at: https://vektorcompute-77d08130.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Security & Encryption

> Institutional-grade data protection for AI workloads.

Processing proprietary enterprise data on a decentralized network requires a paradigm shift in security architecture. Vektor guarantees that node operators can process inference without ever having access to the underlying data or the model weights.

### 256-bit End-to-End Encryption

All API requests submitted to the Vektor network are encrypted at rest and in transit using **AES-256-GCM encryption**.

* **In Transit:** Data traveling between the client and the routing engine is secured via TLS 1.3.
* **At Rest:** Temporary memory states on the GPU during processing are cryptographically isolated.

### Trusted Execution Environments (TEEs)

Vektor mandates that all Node Operators utilize hardware-level Trusted Execution Environments (such as NVIDIA Confidential Computing).

A TEE creates a secure, isolated enclave within the GPU. When an inference request is processed:

1. The encrypted data enters the secure enclave.
2. It is decrypted *only* inside the enclave.
3. The neural network processes the data.
4. The output is re-encrypted before leaving the enclave.

**Result:** The Node Operator, the data center admin, and the Vektor protocol itself cannot physically view, intercept, or copy the data being processed.

### Cryptographic Verification

To prevent malicious nodes from returning falsified or "lazy" inference results to save compute power, Vektor employs **Zero-Knowledge Machine Learning (zkML) proofs**. Nodes must submit a cryptographic proof alongside their inference output, verifying that the computation was executed correctly according to the requested model.