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

WhatsApp has rolled out “Meta AI Incognito Chats,” ensuring conversations are not stored and that officials cannot access them. Experts are concerned this could make AI mishaps difficult to hold accountable. Meta says that, initially, only text will be supported, with conservative safeguards in place.

Meta AI Incognito Chats Debut—Created in Response to People’s Privacy Needs

As generative AI chatbots rapidly become mainstream, chatting with AI has gradually become part of everyday life. However, many questions involve extremely sensitive content—such as providing private financial, personal, health, or work information—and that is when privacy needs arise.

To address this, WhatsApp has introduced “Meta AI Incognito Chats,” claiming it is a brand-new feature that enables completely private conversations with AI.

Based on private processing technology, Meta AI Incognito Chats allow users to have private chats with Meta AI in situations where others cannot see. Users’ messages are processed in a secure environment, and even Meta itself cannot access them.

These conversations are not stored, and by default the system automatically makes messages disappear, giving users a space to think freely and explore ideas without anyone spying on them.

Image source: WhatsApp WhatsApp adds Meta AI Incognito Chats to meet users’ privacy needs

Six Core Technologies Behind Meta AI Incognito Chats

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

1. Confidential computing hardware: Uses AMD CPUs and Nvidia GPUs that support confidential computing to build a trusted execution environment (TEE), ensuring computing tasks are isolated at the hardware level, including Meta or the host operating system, all of which cannot access the data being processed.
2. Authenticated and encrypted communication: The system uses RA-TLS technology for end-to-end encryption, ensuring only the user device and the private processing nodes can decrypt, and verifies the server running an unaltered software stack through hardware attestation.
3. Artifact transparency: To prevent malicious software deployment, the system publishes important components such as binary files and model weights to third-party public transparency logs, so that client-side parties and researchers can verify the authenticity of the code being executed.
4. Secure software: Inside the TEE, the software stack is strengthened through multiple layers of protection, and applications are containerized to limit the attack surface and strictly control data-leakage routes.
5. Anonymous routing: The system uses an anonymous credential service and anonymous HTTP technology to route traffic through third-party relay proxy servers, hiding users’ IP addresses from Meta and de-identifying them to prevent attackers from targeting specific users’ data.
6. Short-lived and stateless data processing: The coordinator and predictor inside the system are designed to be stateless. After processing requests and returning results, the conversation data in memory is discarded to ensure that access to historical records is not retained.

External Concerns: What if Meta AI Is Involved in Unnatural Deaths?

According to a report by the BBC, most AI companies currently store data from chatbots’ usage and use it to train future product model(s). WhatsApp executive Will Cathcart explained that the technology and platform behind WhatsApp’s incognito mode differs from the end-to-end encryption used to protect other messages, but both are equally effective.

Alan Woodward, a cybersecurity expert at the University of Surrey, also noted that deploying a second system poses very low risk to the security WhatsApp already has.

However, external concerns remain that incognito mode could conceal AI failures or abusive behavior. For example, many AI companies—including OpenAI and Google—have previously faced lawsuits related to unnatural deaths.

Woodward believes this could lead to a lack of accountability for AI responses, because automatically disappearing messages cannot be retrieved by users or Meta. That means if someone’s conversation causes harm or death, it will be impossible to find relevant evidence.

In response, Cathcart said that Meta AI Incognito Chats will initially only process text and will not support images for now. Meanwhile, Meta’s safety protections will be conservative and will refuse to answer requests that could be interpreted as harmful or illegal.

In addition, WhatsApp has already blocked other AI chatbots from accessing its system, so for the platform’s billions of users, the only AI they can interact with is Meta’s own products.

Further Reading:
Train AI with Employees! Meta Launches Internal Tracking Tools—Employees’ Mouse Clicks and Keypress Actions Are All Recorded Meta Doubles Down on AI: Zuckerberg Writes Code with Claude, Employees Launch a Token Consumption War to Push KPIs