WhatsApp introduces Meta AI “disappearing” chat, prompting concerns over accountability mechanisms as messages automatically vanish

According to a May 13 announcement on the WhatsApp official website, the social messaging app has launched a “Meta AI Private Processing” feature called “Meta AI Private Processing” for private conversations. It uses six core technologies, including trusted execution environment TEE and RA-TLS end-to-end encryption, to ensure that user messages are processed in a secure environment, and even Meta cannot access the conversation content. After message processing, the messages automatically disappear.

Meta AI Private Processing’s six verification technologies

The six core technologies confirmed in Meta’s technology white paper are:

Confidential computing hardware: Uses an AMD CPU and Nvidia GPU that support confidential computing to build a trusted execution environment (TEE), including Meta and the host operating system, so neither can access data being processed

Authenticated and encrypted communications: Uses RA-TLS technology to provide end-to-end encryption. Hardware attestation verifies whether the server is running an untampered software stack

Artifact transparency: Publishes important components such as binary files and model weights to third-party public transparency logs, allowing clients and researchers to verify the authenticity of the executed code

Secure software: Uses layered defenses and containerizes application programs to limit the attack surface and manage routes for data leakage

Anonymous routing: Uses an anonymous credential service and anti-tracking HTTP technology to route through third-party relay proxy servers, hiding users’ IP addresses from Meta

Transient and stateless data processing: The coordinator and predictor are designed to be stateless; after processing a request, conversation data in memory is discarded to ensure no access to stored historical records

Will Cathcart’s confirmed initial specifications

According to statements from WhatsApp executive Will Cathcart:

Initial limitations: In the initial stage, Meta AI Private Processing only processes text and does not yet support images

Security safeguards: Meta AI’s security mechanisms follow a conservative approach and will refuse to answer requests that could be interpreted as harmful or illegal

Technical positioning: The private processing technology of Private Mode differs from WhatsApp’s existing end-to-end encryption, but it has equivalent effectiveness

Concerns about the accountability mechanism raised by Alan Woodward

S. University cybersecurity expert Alan Woodward told the BBC that the introduction of a second system for Private Mode poses a very low risk to WhatsApp’s existing security. However, he also noted that automatically disappearing messages cannot be retrieved by users or Meta, meaning that if someone’s AI conversation leads to harm or death, no relevant evidence can be found. Woodward said that OpenAI and multiple other AI companies, including Google, have previously faced lawsuits related to deaths under abnormal circumstances, and Private Mode makes accountability in such events harder to carry out.

FAQ

How does Meta AI Private Processing differ from WhatsApp’s existing end-to-end encrypted messages?

According to Will Cathcart’s explanation, the private processing technology used in Meta AI Private Processing and the end-to-end encryption of WhatsApp’s existing messages are different technologies, but both have equivalent effectiveness. The main difference is that after message processing, Private Processing messages automatically disappear, and even Meta itself cannot access them, unlike ordinary messages where the recipient retains possession.

What functions does Meta AI Private Processing currently support, and what does it not support yet?

Based on Will Cathcart’s confirmation, Meta AI Private Processing initially supports only text messages and does not yet support images. Meta AI’s security safeguards follow a conservative approach and will refuse to respond to requests that could be interpreted as harmful or illegal. For the future rollout schedule of other features, WhatsApp has not announced any plans.

Why is Meta AI currently the only AI users can interact with on WhatsApp?

WhatsApp has blocked other AI chatbots from accessing its systems. For the billions of WhatsApp users on the platform, the only AI they can interact with is Meta’s own Meta AI. WhatsApp has not provided detailed explanations for the specific reasons behind this policy.