See sensitive data disappear before it reaches the screen

In this demo, let's take a look at how Island handles sensitive data. The first thing here is a capability Island built into the browser called on-screen data redaction.

So let's look at an example. This is Salesforce. The credentials come from Island's privileged access management tool, and there are two logins here. Both sessions sit in the same Salesforce tenant, on the same case. Two different use cases of the same application. On the left, a privileged user. On the right, a contractor.

On the left, you can see that someone has put credit card data into some Salesforce text fields. This is a privileged user, so maybe that's perfectly fine, because this user needs to process credit card data. But the user on the right is a contractor. They need access to Salesforce. They need access to this case. But they don't need access to the credit card data.

Let's dive into how Island can help us solve this. On this panel, after the data was delivered by Salesforce and decrypted, but before it was painted on the screen, Island scanned it and determined that, based on policy, this user should not see credit card data.

That may be where it stops. Maybe this contractor should never see credit card data. Or policy can allow for the case where a user temporarily needs it. They click into the field, provide the appropriate justification, and only then does that information get temporarily unredacted.

Now watch what happens on copy. If I try moving that data somewhere else, like into an email, it gets restricted. Island knows there's PII in the copy buffer, in this case a credit card number, so it won't let that data move between applications. If I copy only this piece instead, that may be perfectly fine, because there's no PII in the copy. As you can see, when I hit paste, I'm not blocked by Island.

That sets the bar for what on-screen data redaction means. Now I'll move it into an AI use case.

This page has a lot of credit card data on it. Hundreds of credit cards. Island has redacted them by the same policy. This is what it looks like to the user.

This button is the access point to Island's AI browser interactive chat. It's embedded directly in the DOM, the document object model. It's part of the browser experience.

Policy defines which providers and models each user can reach. This user is a contractor, so they get access to five models from three vendors. Users get optionality, access to more than one provider and more than one model, but no more than the ones you want them to have. In this example, I'll use Gemini Pro.

Let's go ahead and type, "What credit cards are on this page?" Now we can let Gemini analyze the question. When Island scans the page and submits the content to the LLM, any data that was redacted is not submitted for summarization. Here I'm seeing an AI summary of the rest of the page, but none of the credit cards were sent. And I can also see it was redacted and restricted by Island's DLP policy.

That's a straightforward use case. The data can't be seen on the screen, so AI can't include it in its summary.

But watch this. If I temporarily unredact a few of these credit cards, then ask that very same question again, "What credit cards are on this page?", Island scans the content of the page. Because it found information of a sensitive nature based on the data protection policy, it restricts the entire page from being sent over to the LLM for summarization.

That's a straightforward use case leveraging Island's AI browser capability. For more ways to boost security and privacy, visit island.io.