Glossary

Trusted Execution Environment (TEE)

An isolated part of the processor that protects code and data while they are being used, and can prove to a remote party what it is running. TEEs are the hardware foundation of confidential computing and of AI inference on data nobody else may see.

General definition

Trusted Execution Environment (TEE) is a secure area of the main processor, specified by the GlobalPlatform industry standards, that runs code in isolation from the rich operating system and all other software on the device. A TEE provides three properties: confidentiality (memory inside cannot be read from outside), integrity (code and data inside cannot be modified from outside) and attestation (the hardware signs a measurement of what was loaded, so a remote party can check it before trusting the environment with keys or data). Encryption protects data at rest and in transit; a TEE is the mechanism that protects data in use, while a CPU is actually computing on it.

Implementations differ in granularity:

  • Arm TrustZone: a secure world alongside the normal world on Arm chips, used in phones for payments, DRM and key storage
  • Intel SGX: process-level enclaves, where an application keeps a protected memory region of its own
  • AMD SEV-SNP, Intel TDX and Arm CCA: VM-level isolation where an entire virtual machine has encrypted memory and is opaque to the hypervisor and the cloud operator
  • AWS Nitro Enclaves: an isolated VM with no storage, network or operator access, attached to a parent EC2 instance
  • GPU TEEs on NVIDIA Hopper and later, extending protection to the memory where model weights and activations sit

The VM-level designs are what cloud providers sell as confidential VMs, and the Confidential Computing Consortium (a Linux Foundation project founded in 2019) defines confidential computing as computation inside a hardware-based, attested TEE. For AI the use case is specific: run inference on sensitive data (patient records, financial transactions, legal documents) so that neither the cloud operator nor an attacker with hypervisor access can read the prompts, the model weights or the outputs; attestation lets the client verify the serving code before sending anything. Limits apply. Side-channel attacks against several TEE designs have been demonstrated, memory encryption and I/O carry a performance cost, and a TEE protects the computation, not the application logic or the data once it leaves. A secure enclave is one form of TEE; the terms are often used interchangeably.

In the Ethora ecosystem

Ethora does not implement a TEE and does not claim attestation of its own code. What it does is keep the trust boundary explicit: the self-hosted stack runs as standard containers, so you can deploy it on a confidential VM from your cloud provider like any other workload, and the AI service connects to any OpenAI-compatible endpoint, so the model can be served from a confidential GPU instance or from a machine inside your own network. In that setup the TEE protects data in use, Ethora’s encryption at rest and TLS in transit cover the other two states, and the platform’s access controls and audit trail govern who can reach the data at all.

Key management stays with the operator: secrets live in your deployment configuration, the React Native SDK keeps credentials in the iOS Keychain and Android Keystore, and the optional end-to-end encrypted rooms hold their keys on the client. For teams whose regulator asks about data in use, a self-hosted Ethora deployment on confidential infrastructure plus a local model is a credible answer; the self-hosted LLM AI agent page covers the model side.

Get started

Chat and AI built for regulated industries

Ethora gives you audit trails, retention controls, a Trust & Safety layer and dedicated deployments in your own cloud. Talk to our team.

Start Free
Free tier available Enterprise SLA No vendor lock-in