TL;DR
Moxie Marlinspike says he is building Confer, a service that applies end-to-end encryption to conversations with AI assistants so only users can access their chat history. He argues conversational AI invites a level of personal disclosure that demands stronger privacy guarantees, and he’s been testing the project privately.
What happened
Moxie Marlinspike, the founder of Signal, published a post describing Confer, an early project that applies end-to-end encryption to AI chat interactions. Marlinspike frames conversational interfaces as uniquely likely to elicit private, unfinished thoughts and reasoning patterns, and says Confer is designed so that those exchanges remain accessible only to the user. He contrasts the medium of chat with search and recalls how expectations around email shifted as people learned corporate servers and third parties could access messages. Marlinspike warns that AI assistants today present a private-chat front while behaving more like an API endpoint feeding a data lake accessible to many actors. He predicts advertising will follow and exploit the fine-grained context revealed in chat. The project is described as nascent; he reports a short testing period with friends and says follow-up technical posts are forthcoming.
Why it matters
- Conversations with AI can expose unpolished thoughts and reasoning patterns that are more revealing than typical posts or searches.
- If chat data is stored or used for training, it could be accessed by companies, partners, adversaries or compelled by law enforcement, altering user expectations of privacy.
- End-to-end encryption for AI chats aims to align a private-looking interface with strong technical protections so users retain exclusive access to their conversations.
- Targeted advertising integrated into assistants could leverage intimate context in ways that change decision-making and persuasion dynamics.
Key facts
- Confer is presented as a service providing end-to-end encryption for AI chat conversations.
- Marlinspike says Confer prevents the service from reading chats, using them for training, or handing them over because only users hold access.
- The author, Moxie Marlinspike, founded Signal and invokes that background when discussing privacy design principles.
- He argues conversational AI encourages people to share incomplete thoughts, context, and reasoning, unlike search which feels transactional.
- Marlinspike compares current AI-chat practices to early email norms, where expectations about privacy shifted after data retention and third-party access became evident.
- The post warns that advertising will eventually target AI assistants and could exploit fine-grained user context and uncertainties.
- Confer is described as an early-stage project; Marlinspike says he has been testing it with friends for a few weeks.
- He signals additional, more technical posts will appear on the project blog in the future.
What to watch next
- Follow-up technical posts on the Confer blog for implementation details and security design decisions.
- Broader testing and any public release plans for Confer (not confirmed in the source).
- The emergence of advertising or commercial models for AI assistants and how they handle conversational data.
Quick glossary
- End-to-end encryption: A method of securing communications so that only the communicating parties can read the messages; intermediaries cannot decrypt the content.
- Large language model (LLM): A type of AI trained on large text datasets to generate or understand human-like language, often used as the core of conversational assistants.
- API endpoint: A network-accessible interface where code can send requests and receive responses from a service or database.
- Data lake: A centralized repository that stores raw or processed data at scale, used for analytics and machine learning.
- Conversational interface: A user interface that allows people to interact with systems through natural language dialogue rather than traditional forms or menus.
Reader FAQ
What is Confer?
An early project described by Moxie Marlinspike that applies end-to-end encryption to AI chat conversations so only the user can access them.
Who is behind Confer?
The project is being developed by Moxie Marlinspike, who founded Signal.
Will Confer read or use my chats for training?
Marlinspike states Confer cannot read chats, use them for training, or hand them over because only users have access.
Is Confer publicly available now?
Not confirmed in the source.
When will Confer launch?
Not confirmed in the source.
Confessions to a data lake Moxie Marlinspike on 23 Dec 2025 I’ve been building Confer: end-to-end encryption for AI chats. With Confer, your conversations are encrypted so that nobody else…
Sources
- Confessions to a Data Lake
- Let's talk about AI and end-to-end encryption
- AI is driving a massive shift towards lake-based data …
- AI Assistants in Encrypted Messaging: Moving Too Fast in the …
Related posts
- Google’s year in review: 8 areas with research breakthroughs in 2025
- Salesforce’s ChatGPT Tie-up Aimed at Preventing Customer Data Leaks
- YC‑backed Keystone seeks founding engineer to build AI error fixer