Introducing ExpressAI: A private-by-design AI platform

Most AI platforms keep your data, so we built one that doesn’t

• Most AI platforms log, retain, or reuse what you submit. ExpressAI was built so you don’t have to accept that trade-off.
• Every interaction is cryptographically isolated from the surrounding infrastructure, so ExpressVPN and model providers can’t access or view your chats.
• Your prompts and files are not used to train models, and nothing is reviewed by humans.
• Your conversation history is protected with zero-access encryption, and you can always set chats to disappear automatically; uploaded files and images are protected using the same zero-access encryption principles, helping ensure stored content remains inaccessible to anyone else.
• ExpressAI runs in the browser and brings multiple models into one interface, alongside built-in web search, a 500 daily credit allowance, and 2GB of secure file storage.
• Independently audited by Cure53, ExpressAI’s architecture has been rigorously tested and validated—so you don’t have to just take our word for it.
  

AI has quietly become the place where people do their most private thinking. Drafts are tested there, half-formed ideas live there, and sensitive material is explored long before it’s ready to be shared.

Yet, most AI tools are built on an assumption that feels increasingly out of step with how people actually use them: that what you type can be seen, stored, or reused.

That trade-off has become normal. Powerful models in exchange for ongoing access to your personal data.

At ExpressVPN, we’ve never accepted that logic. For more than a decade, our work has been guided by a simpler principle: the most reliable way to protect data is to remove access to it entirely. That idea has shaped how we design networks, infrastructure, and security systems…long before AI became part of everyday work.

When AI became part of everyday life, applying the same standard wasn’t optional. We weren’t willing to accept AI systems that could see, retain, or learn from people’s thoughts and work.

So we built ExpressAI—a private-by-design platform that gives you the full power of modern AI without asking you to trade away your privacy.

As Pete Membrey, ExpressVPN Chief Engineering Officer, puts it: “These aren’t just promises, they’re mathematical guarantees.”

That’s because our baseline standards are simple:

• Zero training: Your conversations are never used to train AI models.
• Zero access: No one (not ExpressVPN nor the AI model providers) can access or see your chats and data.
• Zero compromise: You still get access to five powerful models in a clean, focused interface.

ExpressAI begins rolling out on Tuesday, 31 March 2026, starting with new and existing ExpressVPN Pro users.

Privacy that isn’t based on promises alone

Most AI services run on conventional server infrastructure. Prompts may be logged. Outputs may be retained. Over time, conversations become part of pipelines designed to remember and reuse them. 

ExpressAI is built differently.

Every interaction runs inside a confidential computing environment: a secure enclave where data is cryptographically isolated and inaccessible to the host system and infrastructure operators. 

The encryption keys are generated inside the hardware itself, creating cryptographic isolation, which is a state where conversations are mathematically sealed off from everything outside the enclave: cloud providers, model operators, and ExpressVPN.

Access to your data isn’t managed, monitored, or reviewed—it’s simply not possible to do so.

If you choose to keep conversation history, it’s protected with zero-access encryption, so only you can decrypt it. You can also set conversations to Ghost Mode, so they disappear automatically once you’re done. Uploaded files and images are protected using the same principles.

Nothing you share is used to train models. Nothing is reviewed by humans

Independently tested by Cure53

ExpressAI has been independently audited by Cure53, a leading cybersecurity firm known for its rigorous security assessments.

The audit—conducted in February and March 2026—included a penetration test and source code audit of ExpressAI’s frontend and backend components, as well as its cryptography, key management, and underlying infrastructure. 

All identified security vulnerabilities have been addressed and remediated ahead of ExpressAI’s launch.

“In conclusion, the overall architecture and technical stack left a good impression on the testing team,” said Cure53. 

“Cure53 concludes that the product meets its stated privacy objectives by providing modern AI capabilities within confidential computing enclaves, where user interactions are processed in cryptographically isolated contexts.”

This means you can use ExpressAI with greater confidence—knowing its privacy and security claims have been independently tested, not just stated.

You can read the full audit report here.

One place. Multiple models. Full control.

AI models approach problems differently, and ExpressAI doesn’t force you to just choose one. Instead, it brings several high-quality, open-source models into a single interface, so you can choose the right tool for the task or compare responses side by side when judgment matters.

The current lineup includes:

• GPT OSS 120B: A powerful open-weight model by OpenAI for everyday reasoning and writing. Excels at drafting, summarization, Q&A, and general-purpose tasks.
• DeepSeek R1 Distill 32B: Your go-to model for problems that require careful thinking. Excels at multi-step reasoning, logic puzzles, research analysis, and anything that benefits from thinking before answering.
• Qwen2.5-VL 32B: See and understand the world around you. Excels at reading documents, extracting data from images, answering questions about charts, and analyzing diagrams.
• Qwen3.5 35B-A3B: A smart, efficient model for long and complex tasks. Excels at coding, multi-step reasoning, and multilingual tasks.
• Nemotron 12B: A compact powerhouse by NVIDIA for technical work. Excels at solving math problems, generating code, and tackling complex reasoning tasks. 

ExpressAI also features built-in web search, allowing whichever model you choose to fetch real-time information from the web to answer your queries.

Use ExpressAI for your everyday life, or the things you wouldn’t normally trust to AI—reviewing sensitive documents, drafting confidential messages, or working through private ideas. It’s the same way people already use AI, just without exposing what matters.

ExpressAI uses a simple credit system. Sending a prompt to one model uses one credit. Comparing the same prompt across multiple models uses one credit per model. With a generous 500 daily credit allowance, a 50MB upload limit per file, and 2GB of secure vault storage, you have plenty of room to safely analyze heavy documents and complex data. If you reach your daily credit limit, you can still return to existing conversations, and new requests become available again when the allowance resets.

To celebrate the launch, we are offering our most generous package yet: for just $4.87 USD per month, you get access to our entire suite of privacy services. That includes the full power of ExpressAI, our industry-leading VPN, the ExpressKeys password manager, and more—all securely managed under one subscription.

Get ExpressVPN

A different standard

As AI becomes more integrated into everyday life, the stakes around privacy increase. When tools begin to shape how we think, create, and make, the systems behind them matter just as much as the intelligence they provide.

ExpressAI exists because we believe powerful technology shouldn’t come at the cost of your privacy. It’s built on the belief that powerful AI doesn’t require visibility into what people think, write, or upload in order to be useful. 

ExpressAI gives you that space: private by design, and built to let you use modern AI with confidence, not caution. That’s the standard we believe AI should meet, and the one ExpressAI is built to hold.

Try ExpressAI Today