Meta is working on a new technology called Private Processing for WhatsApp that allows users to access artificial intelligence (AI) tools in a private and secure environment. On Tuesday, the Menlo Park-based tech giant shared a first look at the technology and how it is ensuring both data security and transparency at the user level. Notably, the company claims that messages shared with the AI and the responses generated within this cloud-based environment cannot be accessed by anyone (including Meta and WhatsApp) apart from the user and any other person they’re talking to.
The company detailed the vision behind Private Processing and the layered infrastructure that will be used to create this secure environment. It is currently under development, and the company stated that it will soon publish details about some of the components that went into building this technology to enable independent research in this area.
Private Processing is aimed at letting users interact with Meta AI and using features, such as summarising unread chats and generating writing suggestions, without compromising their privacy. This also solves the tech giant’s issue of integrating AI on WhatsApp without raising user concerns about Meta storing user data on its servers and breaking the platform’s end-to-end encryption for messages and other features.
Meta said it is building Private Processing on a trusted execution environment (TEE) infrastructure, a secure part of the cloud that processes data without revealing it to anyone else. For instance, if a user sends a request to the AI to summarise a group’s messages, only the user’s device and the secure processing environment will be able to access this data. Additionally, once the summary has been processed, the information is deleted from the servers.
Coming to technical details, Meta said Private Processing uses advanced cryptographic techniques such as Oblivious HTTP and Remote Attestation to ensure that the user’s identity and data remain hidden. Each request is routed through third-party relays and verified against public ledgers to ensure only approved code is used.
This confidential processing also means that neither Meta nor WhatsApp can access the data entering the cloud environment, either during transit to Private Processing or while processing the data, according to the company.
Meta has also added enforceable guarantees so that if there is an attempt to modify the system, it will automatically trigger system failure. The company also plans to let users and security researchers audit the secure environment and verify the guarantees made.
But these protections only ensure security from potential internal issues. Meta says it is also building safety layers for external threats. As per the post, cyberattackers will not be able to target individual users without compromising the entire system. Further, since the data is only stored temporarily on the server, attackers will not be able to retrieve older data even if they can pull off a server-wide attack.
In the coming weeks, Meta said it will release more details, including technical papers and bug bounty expansions. While AI-powered features like message summarisation will be part of Private Processing’s initial offering, the company plans to add several other use cases in the future. For more information, readers can find the entire blog post here.
