WhatsApp推Meta AI 無痕對話：號稱連Meta都看不了，但有一大隱憂

WhatsApp Launches “Meta AI Incognito Chat,” Ensuring Conversations Are Not Stored and Are Inaccessible to Officials. Experts Concern That This Could Make AI Accountability Difficult, with Meta Stating It Initially Supports Only Text and Uses Conservative Protections.

Meta AI Incognito Chat Debuts, Born from Privacy Demands

As generative AI chatbots rapidly become widespread, talking with AI has gradually become part of daily life for the public. However, many questions involve highly sensitive content, such as providing private financial, personal, health, or work information, creating privacy needs.

In response, WhatsApp has launched “Meta AI Incognito Chat,” claiming to be a new feature for completely private conversations with AI.

Meta AI Incognito Chat is based on private processing technology, allowing users to have confidential chats with Meta AI without others seeing. User messages are processed in a secure environment, inaccessible even to Meta itself.

These conversations are not stored, and the system defaults to message auto-deletion, providing users a space to think freely and explore ideas without anyone spying.

Image source: WhatsApp WhatsApp introduces Meta AI Incognito Chat, born from privacy needs

Six Core Technologies of Meta AI Incognito Chat

According to Meta’s technical white paper, Meta AI Incognito Chat primarily combines the following six core technologies to ensure the privacy and security of conversation data:

1. Confidential Computing Hardware: Uses AMD CPUs and Nvidia GPUs supporting confidential computing to establish a Trusted Execution Environment (TEE), ensuring computations are isolated at the hardware level, including Meta or host operating systems, which cannot access the data being processed.
2. Authenticated and Encrypted Communication: The system uses RA-TLS technology to provide end-to-end encryption, ensuring only the user device and private processing nodes can decrypt, and hardware authentication verifies whether the server runs unaltered software stacks.
3. Artifact Transparency: To prevent malicious software deployment, the system publishes critical components such as binary files and model weights to third-party public transparency logs, allowing clients and researchers to verify the authenticity of the code being executed.
4. Secure Software: Inside the TEE, the software stack is reinforced with multiple layers of protection, and applications are containerized to limit attack surfaces and strictly control data leakage pathways.
5. Anonymous Routing: The system employs anonymous credential services and onion routing techniques, routing through third-party relay servers, hiding user IP addresses from Meta and de-identifying users to prevent attackers from targeting specific individuals’ data.
6. Ephemeral and Stateless Data Processing: The coordinator and predictor within the system are stateless; after processing requests and returning results, conversation data in memory is discarded, ensuring no access to historical records.

External Concerns: What Happens if Meta AI Involves Unnatural Death

According to BBC reports, most AI companies currently store chat data and use it to train future models. WhatsApp executive Will Cathcart explained that the technology and platform behind WhatsApp Incognito Mode differ from end-to-end encryption used for other messages, but both are equally effective.

Surrey University cybersecurity expert Alan Woodward also noted that implementing a second system poses very low risks to WhatsApp’s existing security.

However, there are concerns that incognito mode could mask AI failures or abuses. Multiple AI companies, including OpenAI and Google, have previously faced lawsuits related to unnatural death cases.

Woodward believes this could lead to AI responses lacking accountability, as messages that automatically disappear cannot be accessed by users or Meta. If someone’s conversation results in harm or death, relevant evidence may be impossible to find.

Cathcart responded that Meta AI Incognito Chat will initially only support text and not images. Meanwhile, Meta’s safety protections will be conservative, refusing to answer requests that could be interpreted as harmful or illegal.

Additionally, WhatsApp has blocked other AI chatbots from accessing its system, so the only AI interactions available to its hundreds of millions of users are with Meta’s own products.

Further reading:
Training AI with Employees! Meta Launches Internal Tracking Tools, Recording Employee Mouse and Keyboard ActionsMeta Doubles Down on AI: Zuckerberg Uses Claude to Write Code, Employees Launch Token Consumption War to Hit KPIs