Gartner's current advice is to block AI browsers entirely — because they work by sending as much data as possible to the LLM vendor. Island's AI browser is built differently. Organizations can bring their own model, tailor the experience by user role and context, and enforce data boundaries in real time — blocking AI from accessing sensitive information while still letting it work where it's safe to do so. The flexibility to choose the right model for the right team, without sacrificing control, is what makes it enterprise-ready.

Read on for a transcript of the AI Browser demo video.

Now, the AI Enterprise Browser. This isn't a new product. Organizations can choose whether to enable the AI capabilities. Gartner recently stated that cybersecurity teams must block AI browsers for now, because for these browsers to work, they send as much data as possible to the LLM vendor for better context and results. That's the problem.

What we did is take the main pillars of an AI browser and make them enterprise-ready. If you roll out OpenAI's browser, you're married to OpenAI's LLM. If you use Chrome, you're married to Gemini. But what if Claude is a better model for your engineers? This isn't like the early search engine days where there's one primary truth. Organizations need flexibility.

In 2025 alone, there were periods where Grok excelled at sentiment analysis due to its Twitter integration, Gemini outperformed others on certain tasks, and Anthropic became the de facto choice for software engineers. Organizations need to bring their own model. They need data protection controls: what can users type in, what can be shared with the cloud. They need enterprise context: who the user is, their job function, what business applications they connect to.

So we implemented enterprise chat, an enterprise agent within the browser, enterprise context, and many more capabilities. And we made it beautiful. We worked hard on the edges, the latency, the animations — and we're improving it every day.

Here are the new AI browser features. We have a new Omnibar at the top and an embedded chat — which can be any chat: our enterprise chat, Copilot, Gemini, Grok, Claude, ChatGPT, you name it.

What makes this unique is the ability to customize and tailor the experience. Think about doctors — AI could make them significantly more efficient and faster. But the way AI browsers work today makes that nearly impossible to implement safely. What I'm about to show you is unique for a few reasons.

First, customization. The browser is branded, and the entire experience is branded. We can suggest smart actions based on the context of the website and the identity and role of the end user — because a doctor and a nurse may need two different things when viewing the same medical record.

When I ask the system to analyze the vitals on this page, Copilot attempts to access the page content. But Island acts as an AI web application firewall and blocks the request because it violates the organization's DLP and data boundary policies.

Here's what makes it really interesting: we can distinguish between situations where sensitive information is present and where it isn't. We can mask the data on the page — the name, ID number, and SSN are no longer present in the DOM. When I ask again, Copilot tries to access the page content, we evaluate it against our data policy, and because there's no sensitive information, it's able to proceed.