WhatsApp has unveiled a new technology called Private Processing designed to enable AI capabilities while preserving user privacy. This feature, expected to roll out in the coming weeks, allows users to access AI functions like message summarization and editing assistance without compromising WhatsApp’s core privacy promise.

## How Private Processing Works

The system operates within a secure environment called the confidential virtual machine (CVM), ensuring that no third party—including Meta and WhatsApp—can access users’ messages. When a user initiates an AI request, the system:

1. Verifies legitimate WhatsApp clients through anonymous credentials
2. Establishes an Oblivious HTTP connection via a third-party relay that masks the user’s IP address
3. Creates a secure session between the device and the Trusted Execution Environment (TEE)
4. Processes data in the CVM using encryption that only the user’s device and the Private Processing server can decrypt

## Key Privacy Principles

Private Processing is built on several foundational principles:

– **Confidential processing**: Messages are processed in a secure environment
– **Enforceable guarantees**: The system fails or becomes publicly discoverable if tampered with
– **Verifiable transparency**: Users and researchers can audit system behavior
– **Non-targetability**: Prevents targeting specific users without compromising the entire security architecture
– **Stateless processing**: Messages aren’t retained after processing, preventing historical data recovery

Meta acknowledges potential threats from compromised insiders, supply chain risks, and malicious users, but emphasizes its defense-in-depth approach to minimize vulnerabilities. The company plans to publish third-party logs of CVM binary digests and images to enable external security verification.

This approach parallels Apple’s Private Cloud Compute system, which similarly uses OHTTP relay and sandboxed environments for confidential AI processing.