Skip to main content

​
Overview

A Trusted Execution Environment (TEE) leverages hardware to provide secure and efficient encryption, ensuring that sensitive data remains protected. TEEs operate independently from the main operating system, providing a protected execution space for sensitive operations. In the context of Omo Protocol, TEEs are used to obfuscate agent keys and secure ML model execution, enhancing security through the Oasis Sapphire Layer. Key benefits of TEEs include:
  • Enhanced security through hardware-based isolation
  • Efficient encryption for faster processing
  • Reduced attack surface by isolating critical operations
  • Improved compliance with data protection regulations
Trusted Execution Environment (TEE) Architecture

Trusted Execution Environment (TEE) Architecture

​
How TEE Cryptography Works

TEE cryptography relies on hardware-based isolation and encryption:
1

Encryption and Loading

Code and data are encrypted and loaded into the TEE’s secure area.
2

Access Control

Only authorized code can access the protected area.
3

Secure Computation

The TEE performs computations on the encrypted data.
4

Secure Output

Results are securely output, maintaining confidentiality throughout the process.

​
TEEs in Omo: Agent Key Security and teeML

At Omo, we leverage TEEs for two primary purposes: enhancing agent key security and implementing our novel teeML approach.

​
Agent Key Security

We use TEEs to obfuscate agent private keys, providing an additional layer of protection:
  • Key Isolation: Agent private keys are stored within the TEE’s secure enclave, isolating them from the main operating system.
  • Access Control: Only authorized processes within the TEE can access the private keys, preventing unauthorized use.
  • Tamper Resistance: Even if the host system is compromised, the TEE’s hardware-based security prevents unauthorized access to agent credentials.

​
teeML: Secure Machine Learning

Our teeML approach combines TEEs with machine learning capabilities to ensure both security and efficiency:
  • Model Storage: ML model files are securely stored within the TEE, protecting them from unauthorized access or tampering.
  • Secure Inference: When an agent needs to make an inference, it calls the model stored in the TEE, ensuring that both the model and the inference process remain confidential.
  • Efficient Verification: TEEs enable verification of inferences from larger ML models at a fraction of the cost, offering fast validity times and cost-effective processing for even the largest models.
This implementation allows us to leverage the power of machine learning while maintaining high standards of security and privacy throughout our protocol. The full potential of teeML is being explored by the FOMO Labs team and our respective partners.

​
Risks of Using TEEs

  • Most TEEs have experienced vulnerabilities over time
  • Confidentiality breaches are more common than integrity breaches
  • Complex architecture and closed-source development contribute to vulnerabilities
  • Strong adversary models make TEEs attractive targets for attacks
  • TEEs are vulnerable to various side-channel attacks - Attackers can potentially extract sensitive information through indirect means - Both service providers and remote attackers pose threats to TEE security
  • Reliance on central issuers (e.g., Intel for SGX) introduces trust concerns
  • Users must trust the hardware manufacturer and their implementation - Inverting trust assumptions may be necessary in some cases
  • In blockchain applications, TEE usage can lead to centralization
  • Limited number of TEE providers may create bottlenecks or single points of failure
  • Over-reliance on TEEs can compromise the decentralization ethos of blockchain systems

​
Mitigating TEE Risks

To address the inherent risks of TEEs, we employ two primary strategies:

​
MPC Networks

Multi-Party Computation (MPC) networks complement TEEs by distributing trust and computation across multiple parties:
  • Reduces reliance on a single TEE or provider
  • Enhances privacy through distributed computation
  • Improves overall system resilience against attacks
In our implementation, we use MPC to create a network of TEEs, ensuring that no single point of failure can compromise the entire system.

​
Combined Cryptographic Approaches

We leverage a combination of cryptographic techniques alongside TEEs:
  • Zero-Knowledge Proofs (ZKPs):
    • Allow verification of computations without revealing sensitive data
    • Complement TEEs by providing an additional layer of privacy and integrity
  • Optimistic Fraud Proofs:
    • Enable efficient verification of large computations
    • Work in tandem with TEEs to ensure the correctness of operations at scale
By combining these approaches, we create a robust, multi-layered security architecture that addresses the inherent risks of TEEs while leveraging their strengths for enhanced security and ML capabilities in the Omo Protocol ecosystem.