The Guy Who Built Signal's Encryption Is Now Securing Your Meta AI Conversations

If you've ever had a private conversation with an AI chatbot and wondered who else might be reading it — this one's for you.

Moxie Marlinspike, the cryptographer who built Signal's end-to-end encryption protocol (the same tech that already secures your WhatsApp messages), has created a privacy-focused AI platform called Confer. And now he's integrating that encryption technology directly into Meta AI.

The idea is simple in theory, wild in execution: your conversations with Meta's AI assistant across WhatsApp, Messenger, and Instagram would be encrypted so that not even Meta can read them.

Think about what AI chatbots know about you. People use them as therapists, financial advisors, writing partners, and confessional booths. Every one of those conversations currently sits on a company's servers, readable by employees, vulnerable to breaches, and potentially used to train future models or target ads. Marlinspike's argument is that if a chatbot feels like a private conversation, it should actually work like one.

The technical approach uses a combination of homomorphic encryption (which lets AI process data without ever decrypting it) and Trusted Execution Environments — hardware-enforced isolated computing spaces where data gets processed securely. Your prompts get encrypted on your device, stay encrypted during processing, and the response comes back encrypted too.

This is a big deal because it solves a problem nobody else has seriously tackled. Every major AI company — OpenAI, Google, Anthropic — processes your queries in plaintext on their servers. They have to, because that's how the models work. Confer's approach is computationally expensive and technically complex, but if it delivers, it's a genuine breakthrough.

There's an ironic twist worth noting though. While Meta is adding encryption to its AI conversations, the company is simultaneously removing end-to-end encryption from Instagram DMs starting May 2026, citing child safety concerns. So your conversations with a robot get privacy protection, but your conversations with actual humans might not. Make that make sense.

Marlinspike launched Confer as a standalone ChatGPT alternative in December 2025. The Meta AI integration is happening through 2026, though specific launch dates haven't been announced.

For the billions of people who talk to Meta's AI every day across WhatsApp and Instagram, this could be transformative. The question is whether Meta will actually commit to it fully, or whether the encryption becomes a marketing talking point that doesn't survive contact with the company's ad-targeting business model.

Privacy skeptics have every reason to be cautious. But having the guy who literally wrote the encryption standard that protects billions of messages involved? That's about as credible as it gets.

What do you think — does encrypted AI actually matter to you, or have we already accepted that everything we type gets read by someone?