The Creator of Signal Is Assisting in Securing Meta AI

Moxie Marlinspike, the privacy advocate behind the secure messaging app Signal and its popular open source encryption protocol, announced this week that his privacy-centric AI platform, Confer, will begin integrating its technology into Meta’s AI systems.

Billions of chat messages exchanged daily via Signal, Meta’s WhatsApp, and Apple’s Messages benefit from end-to-end encryption. This feature, which prevents tech companies and anyone aside from the sender and recipient from viewing your messages, has gained mainstream acceptance over the past decade. However, as generative AI platforms surge in usage, users are now sending billions of messages daily to AI chatbots that lack end-to-end encryption—allowing AI companies easy access to the content shared.

This is intentional, as platforms typically aim to utilize as much user data as possible for training their AI models, making it challenging to opt out of data collection. Yet, as AI agents and chatbots advance in capability, some technologists and companies are advocating for more privacy-focused and constrained systems.

“With LLMs becoming increasingly capable, we should anticipate more data flowing into them,” Marlinspike noted in a brief blog post about his partnership with Meta published on Tuesday. “Currently, none of that data remains private. It is accessible to AI companies, their staff, hackers, legal requests, and government entities. As with any unencrypted data, it will inevitably fall into the wrong hands.”

Marlinspike stated he intends to “integrate Confer’s privacy technology as the foundation for Meta AI.” He highlighted that Confer, which launched at the start of this year, will continue to function independently from Meta. According to Marlinspike, the initiative aims to deliver technology that combines the full potential of AI with the complete privacy of an encrypted conversation.

In 2016, Marlinspike collaborated with WhatsApp, owned by Meta, to implement end-to-end encryption for more than a billion accounts at once. Over the past year, WhatsApp has introduced a Meta AI chatbot within its app, which does not have the same protections as individual chats.

“People engage with AI in deeply personal ways that necessitate access to confidential information,” WhatsApp head Will Cathcart commented on Wednesday via the social media platform X regarding the collaboration with Confer. “It’s vital that we construct this technology in a manner that empowers users to maintain privacy.”

The embrace of encrypted AI is still in its nascent stages. The cryptographic methods used for end-to-end encryption in conventional digital communication aren’t easily applicable to generative AI protections. Confer is a new initiative, and Marlinspike’s blog post didn’t offer detailed information on how the collaboration with Meta will function or the specific goals for integration.

Neither Marlinspike nor Meta offered further comment to WIRED prior to publication.

Mallory Knodel, a cryptography researcher at New York University, remarked that it would be “excellent for users of chatbots that utilize Meta AI to enjoy confidentiality and privacy within that interaction.” Importantly, this means Meta would not access AI chat data for training purposes, according to Knodel, who, along with colleagues, recently published research on end-to-end encryption and AI. “I sincerely hope more AI chatbots adopt this methodology.”

Knodel’s preliminary evaluations of Confer indicate that while the platform is not flawless, it serves as a significant example of how to construct a private AI chatbot.