For years, ExpressVPN has been the company you turned to when you wanted your browsing to disappear from prying eyes—now it wants to do the same for your AI chats. With the launch of ExpressAI, the privacy-focused VPN giant is stepping directly into the AI race, but with a clear message: powerful models don’t need access to your thoughts to be useful.

Most popular chatbots today run on an implicit bargain: you get great answers, and in return, your prompts might be logged, retained, or folded back into future training data. ExpressVPN is very openly saying “no thanks” to that bargain. ExpressAI is built on confidential computing, meaning your conversations run inside secure hardware enclaves that even cloud providers and ExpressVPN themselves supposedly can’t peek into. The encryption keys live inside this hardware, and data is decrypted only inside the enclave; in other words, your chats are mathematically sealed off from the rest of the infrastructure instead of being protected by policy alone.

The privacy story doesn’t stop there. ExpressAI leans on three core promises: “zero training,” “zero access,” and “zero compromise.” Zero training means your prompts and files aren’t used to train any of the underlying models, now or later. Zero access means neither ExpressVPN nor its model partners can read your chats or attached documents, because they’re encrypted and processed inside those secure enclaves. And zero compromise is the pitch that you still get access to multiple capable models and modern AI features without giving up your privacy to get them.

Because privacy claims in tech have been overused to the point of cliché, ExpressVPN pushed ExpressAI through an external audit before launch. Cybersecurity firm Cure53 ran penetration tests and a source-code review across the frontend, backend, cryptography, and infrastructure in February and March 2026, then concluded that the system does what it says on the tin: user interactions are processed in cryptographically isolated contexts that match the platform’s stated privacy objectives. ExpressVPN says all issues discovered during the audit were fixed before ExpressAI rolled out to users.

On the surface, ExpressAI looks familiar if you’ve used tools like ChatGPT or Gemini. You can use it to summarize long documents, draft emails, analyze images, debug code, or bounce around half-formed ideas, all in a browser interface that feels closer to a clean writing space than a bloated productivity suite. Under the hood, though, it’s less of a single AI and more of a curated collection of models wired into one privacy‑first wrapper.

At launch, ExpressAI supports five different models from four providers. GPT OSS 120B, an open-weight model by OpenAI, handles general reasoning and writing tasks like drafting, summarization, and Q&A. DeepSeek R1 Distill 32B is the “think slowly” option, aimed at multi-step reasoning, logic puzzles, and deeper analysis where you want the model to reason before answering. Qwen2.5-VL 32B focuses on vision and documents—it can read PDFs, parse charts, and answer questions about images or diagrams. Qwen3.5 35B-A3B is optimized for long, complex work and multilingual scenarios, including coding. Finally, NVIDIA’s Nemotron 12B is the compact specialist for math, code generation, and technical problem‑solving.

One of the more practical touches is that you’re not locked into a single model. ExpressAI lets you send the same prompt to multiple models and compare the answers side by side, which is especially useful for tricky questions where nuance matters or you simply want a second opinion before acting on anything. There’s also a built‑in web search, so your chosen model can pull in real‑time information when you need current facts, not just what the model already “knows.”

Privacy extends into how ExpressAI handles your history and files. If you keep your conversation history, it’s protected with what ExpressVPN calls “zero-access encryption,” where only you hold the keys needed to decrypt your chats. For more sensitive work, there’s Ghost Mode: turn it on, and conversations automatically disappear once you’re done, leaving no stored record on ExpressVPN’s side. The same encryption principles apply to uploaded files and images, which are stored in a secure vault behind a password you control.

Unlike some competitors that quietly encourage heavy usage and then upsell you, ExpressAI is upfront about a credit-based system, but the allowance is fairly generous. Every prompt you send to a model costs one credit; send that same prompt to two models for comparison, and that’s two credits. ExpressVPN is starting with 500 credits per day, a 50MB limit per file upload, and 2GB of encrypted storage—more than enough for everyday work, code snippets, and hefty PDFs. If you hit the daily cap, you can still revisit old conversations, and new credits show up again after the reset.

There’s also a clear commercial strategy here. ExpressAI isn’t a standalone app you buy separately; it’s part of ExpressVPN’s broader “privacy suite.” To mark the launch, ExpressVPN is offering a package starting at $4.87 per month that bundles ExpressAI, the company’s core VPN service, the ExpressKeys password manager, and other tools under one subscription. For ExpressVPN Pro customers, ExpressAI begins rolling out first, with broader access expected to follow.

For everyday users, the pitch is pretty straightforward: use AI the way you already do—drafting job applications, reviewing contracts, planning travel, even journaling your thoughts—without worrying that those private snippets are being stored or data-mined. For more privacy-sensitive groups—journalists, lawyers, policy teams, or anyone handling confidential client information—the confidential-computing angle and the third-party audit may be the real differentiators.

Of course, ExpressAI doesn’t magically solve all AI concerns. Like any AI system, its output can still be wrong or biased, and users still have to bring their own judgment to whatever it says. But ExpressVPN is taking a clear position in a space dominated by data-hungry platforms: your prompts are your business, not your provider’s training data. In an AI moment defined by “more data, bigger models,” ExpressAI is gambling that the next competitive edge might actually be knowing as little about you as possible.