Meta introduced Private Processing, an optional new feature aimed at allowing WhatsApp users to process messages with AI in a private, secure cloud environment. Meta stated this means that neither WhatsApp nor Meta any third party is able to access the messages, preserving end-to-end encryption.
The news emphasized how AI has redefined technology engagement through the automation of activities and insights on data. However, traditional AI processing, which relies on server-based large language models, often requires providers to see user requests.
This can challenge privacy, especially for sensitive messages. Meta stated that the Private Processing tackles this issue by supporting AI functions, such as summarizing messages or offering writing assistance, while upholding WhatsApp’s commitment to privacy.
Meta defined three guiding principles for Private Processing:
Optionality: Utilizing AI features, including Private Processing, is completely optional.
Transparency: The firm will clearly state when Private Processing is active.
User Control: Users can disable AI features in sensitive chats with WhatsApp's Advanced Chat Privacy feature.
Security Measures:
Meta created a threat model to determine the risks, with the following priorities:
Assets: Defending against message content (delivered or authored) and system components such as the CVM, hardware, and encryption keys.
Threat Actors: Malicious insiders, third-party suppliers, or end-users attacking others.
Threat Scenarios: Attacks could include exploiting weaknesses, extracting information from CVMs, or tampering with hardware.
Meta defended against these using:
System Software: No remote shell access, code isolation, auditable code modifications, and secure build procedures.
System Hardware: Confidential virtualization on CPU-based and Computer mode GPUs to prevent host or physical attacks.
Defense-in-Depth: OHTTP relays, encrypted DRAM, and physical security of data centers to avert targeted attacks.
