What we learned about TEE security from auditing WhatsApp's Private Inference (blog.trailofbits.com)

• Analysis of TEE trust boundaries in public cloud environments
• Evaluation of remote attestation flaws in WhatsApp inference
• Risks of hardware-based side-channel attacks on private AI

"A detailed technical review of WhatsApp's integration of Trusted Execution Environments (TEEs) for private inference. The audit assesses how effectively TEEs protect user data during machine learning tasks, highlighting significant gaps in remote attestation and potential hardware-level vulnerabilities. By analyzing the trust boundary between cloud service providers and end-users, the study underscores the complexities of deploying privacy-preserving AI in centralized infrastructures, offering a critical perspective on the trade-offs between verifiable security and high-performance cloud computing."