Signal founder Moxie Marlinspike launches Confer, an encrypted AI chatbot alternative to ChatGPT

Signal Founder Tackles AI Privacy with Confer Launch

Moxie Marlinspike, the engineer behind Signal Messenger, has unveiled Confer, a privacy-focused AI service designed to address mounting concerns about user data privacy in artificial intelligence platforms 1

1

. Launched in December, this alternative to ChatGPT represents an attempt to apply the same end-to-end encryption principles that made Signal trusted by millions to the world of AI chatbots 2

2

. The open source platform encrypts both user prompts and AI responses, ensuring that conversations remain inaccessible to platform operators, hackers, law enforcement, or advertisers 1

1

.

The timing reflects growing unease about how mainstream AI platforms handle sensitive information. Last May, a court ordered OpenAI to preserve all ChatGPT user logs—including deleted chats and sensitive conversations from API business offerings—highlighting how user data remains vulnerable to legal subpoenas 1

1

. Sam Altman, CEO of OpenAI, acknowledged that even psychotherapy sessions on the platform may not stay private under such rulings. With OpenAI already testing advertising, concerns intensify about the same data collection practices that fuel Facebook and Google creeping into chatbot conversations 2

2

.

How Confer Achieves Truly Private AI Conversations

Confer's architecture combines multiple security layers to deliver on its privacy promise. The AI chatbot uses passkeys—an industry-wide standard that generates a 32-byte encryption keypair unique to each service 1

1

. The public key goes to the server, while the private key remains securely stored on user devices in protected hardware that even physical attackers cannot access. Users authenticate through fingerprints, face scans, or device unlock PINs—all of which stay locally on their devices 1

1

.

The platform leverages WebAuthn and passkey encryption to encrypt messages before they leave user devices 3

3

. On the server side, all processing happens inside Trusted Execution Environments (TEEs)—hardware-enforced isolated spaces where the host machine provides CPU, memory, and power but cannot access the TEE's memory or execution state 4

4

. Data and conversations originating from users and resulting responses from large language models are encrypted within these TEEs, preventing even server administrators from viewing or tampering with them 1

1

.

Confer employs remote attestation, allowing anyone to verify what code runs on its servers 3

3

. Each release is digitally signed and published to a transparency log, offering developers and organizations a way to assess how their data is handled 5

5

. This confidential computing approach means prompts are encrypted on a user's device, sent to Confer's servers in encrypted form, then decrypted only in the secure environment to generate a response 4

4

.

Why AI Privacy Matters More Than Search Privacy

Marlinspike argues that interactions with AI assistants differ fundamentally from traditional web searches. Users treat dialogue as intimate conversation, sharing thoughts, fears, transgressions, business dealings, and secrets as if AI assistants are trusted confidants or personal journals 1

1

. This contrasts sharply with web search queries, which typically follow a transactional model of keywords in and links out. He describes AI use as confessing into a "data lake" 1

1

.

Data privacy expert Em calls AI assistants the "archnemesis" of data privacy because their utility relies on assembling massive amounts of data from myriad sources. "AI models are inherent data collectors," she explained. "They rely on large data collection for training, improvements, operations, and customizations. More often than not, this data is collected without clear and informed consent" 1

1

. Research from the National Cybersecurity Alliance found that over 40% of workers have shared sensitive information with AI 3

3

.

Marlinspike warns that large language models represent the first major tech medium that "actively invites confession." As people chat with these systems, they reveal thinking patterns and uncertainties that could be exploited by advertisers to sell products or influence behavior 4

4

. "It's a form of technology that actively invites confession," Marlinspike says. "Chat interfaces like ChatGPT know more about people than any other technology before. When you combine that with advertising, it's like someone paying your therapist to convince you to buy something" 2

2

.

Pricing, Availability, and Growing Adoption

Confer offers a free tier limited to 20 messages per day and five active chats 2

2

. Users willing to pay $35-per-month gain unlimited access, along with more advanced models and personalization features 2

2

. That pricing sits notably higher than ChatGPT's Plus plan, reflecting the infrastructure costs of maintaining encryption and Trusted Execution Environments.

The platform works best on Mac and Android devices, though Windows and Linux users can access it through compatible password managers 3

3

. Confer uses different AI models for different tasks, sourced from the open source community, with more advanced modeling available in premium subscriptions 3

3

. The platform is in active development, with recent upgrades including the option to import ChatGPT or Claude conversations, and an iOS app on the horizon 3

3

.

Marlinspike reported rapid growth following the launch, noting on X that traffic has been climbing through early Signal milestones like 1 message per second, 10 per second, and 100 per second 3

3

. He shared that it's been "really interesting and encouraging and amazing to hear stories from people who have used Confer and had life-changing conversations," in part because they haven't felt free to include certain information in conversations with ChatGPT but can in Confer's encrypted environment 1

1

.

What This Means for AI's Privacy Future

Confer inverts the privacy model that dominates AI platforms. While ChatGPT, Gemini, and Meta AI provide opt-out toggles for chat history and training data usage, the default state remains surveillance-oriented, with opting out being the user's responsibility 5

5

. Confer makes privacy the default, baked into the architecture rather than offered as an afterthought 5

5

.

Organizations like schools and hospitals exploring AI tools might find appeal in platforms that guarantee confidentiality by design 5

5

. Signal was founded in 2014 around similar principles, and its open source encrypted messaging protocol was adopted by Meta's WhatsApp just a few years later 4

4

. Whether tech giants eventually adopt Confer's technology remains to be seen, but the platform raises awareness about what privacy-by-design could look like in AI. "The core idea is that your conversations with an AI assistant should be as private as your conversations with a person," Marlinspike states. "Not because you're doing something wrong, but because privacy is what lets you think freely" 3

3

.