Demo Video

See every AI tool running across your endpoints

https://vimeo.com/1248352532?share=copy&fl=sv&fe=ci

Watch how Island inventories the AI applications, MCP servers, skills, hooks, code packages, and IDE extensions on every device — then scores each one for risk and explains why. You'll also see how Island audits every AI prompt and response, with the full conversation stored in your own S3 bucket or Azure Blob.

‍

Transcript

In this demo, let's take a look at the raw data in the AI Protect dashboard. Here I can see a list of every AI application that has been discovered, either in the browser, on the desktop, or in transit.

Furthermore, I can see every MCP server that's also been discovered, which applications are currently connected through it, how it's communicating and whether it's local or remote, and the risk score of that MCP along with the vulnerabilities it may have. Island runs its own assessment of the risk of these MCP servers, and these are the findings that explain that risk.

I can also see a list for AI skills: a list of all the skills in use and which application is leveraging them. The risk score here fleshes out the entire risk posture. A skill is nothing more than a text file. It's a Markdown file. Island ingests the entire skill file and runs an AI risk assessment across it.

For example, Island can tell you exactly why this skill is risky. This skill can potentially handle sensitive data, it's at risk for prompt injection, it can automate the browser, it has network access, and it can read and write files. Then every device that has it installed, and where it was found on that device, is also listed.

Moving on, I can see a list for AI hooks. AI hooks are hooks into the various AI applications at the desktop level. Just like MCP servers and AI skills, I can see which application is being hooked, a risk score, the hook details, and the assessment.

Looking at code packages, Island can tell you every code package being installed and its origin. A script written by an AI tool that goes out to download a malicious package from an untrusted source would be seen here, and could have been prevented by the right policy in the web gateway.

And for developers integrating AI coding assistants into their IDEs, every IDE that's been discovered with those applications installed can be viewed here, along with a breakdown of every version on each device and the users running each one.

This capability is all about visibility and audit. It covers the individual events generated for every AI interaction, whether it came from a user, an application, or an API.

If I open any one of these AI prompts and look at the full conversation, the entire conversational history of that user's interaction with a generative AI tool is displayed. Scroll to the very top where it started, and there's the initial AI prompt, the initial AI response, the secondary prompt, the secondary response. All of it is fully audited to examine.

Island doesn't hold this data in its own infrastructure. Customers provide a file location, typically an S3 bucket or an Azure Blob, and it's all stored there. As long as you're maintaining that location, for things like CMMC compliance and governance, Island is completely out of scope. That applies to all AI events, interactive and generative.

To learn more about Island's Enterprise AI, visit island.io.

‍