> ## Documentation Index > Fetch the complete documentation index at: https://docs.contentignite.com/llms.txt > Use this file to discover all available pages before exploring further. # AI and your data > How Spark and AI summaries keep your data private and secure Fusion's AI features, [Spark](https://docs.contentignite.com/ai/spark.md) and [AI summaries](https://docs.contentignite.com/ai/summaries.md), work with your live Fusion data, so it's only natural to want to know how that data is looked after. Here's how we keep it private, secure and in your hands. ## The short version * **Your organisation's data stays yours.** Fusion filters everything by organisation before the AI sees it, so other organisations' data is off limits, and yours is off limits to them. * **Spark sees what you see.** It uses your own login, so it can only reach data you could already see in Fusion. * **Spark is read-only.** Right now, it can look things up and explain them, but it can't make changes to your account. * **Your Spark chats are private.** Other users can't open your conversations. * **Your data isn't used to train AI models.** Not by us, and not by the provider of the AI model behind Fusion's AI. ## How Fusion's AI reaches your data Fusion's AI only gets the data it's allowed to see, so there's nothing else for it to share. Whenever Spark answers a question or a summary is written, the data is looked up through the same checks that decide what can be seen everywhere else in Fusion. That means: * **One organisation at a time.** Spark answers for the organisation you're working in, and summaries are written for your organisation. Switch organisation in Fusion and Spark switches with you. * **Fusion is in charge, not the AI.** The AI doesn't decide what it's allowed to see. Fusion checks each request and only hands back data that the login has access to, and what's typed into a chat doesn't change that. Fusion's AI is also instructed not to share passwords, keys or details of how our systems work, and to treat any instructions hidden in your data, such as a publisher name that tries to tell it what to do, as plain text. These instructions are a helpful extra, but the main protection comes from the AI only having your organisation's data to work with. ## Spark ### Spark sees what you see Spark looks up every answer using your own login, so your role still applies. If your role can't see something in Fusion, Spark can't see it either. Organisation admins can manage roles from the [Organisation page](https://docs.contentignite.com/organisations.md#user-roles). ### Read-only, for now Right now, Spark can look things up and explain them, but it can't create, change or delete anything in your account. This isn't just an instruction we give Spark. It can only reach the parts of Fusion that read data, so making changes is out of its reach. If you ask Spark to change something, it will point you to the right place in Fusion and walk you through the steps so you can do it yourself. ### Your conversations Each chat belongs to the person who started it, in the organisation they were working in. Other users can't open your conversations, including people in your own organisation, unless you choose to share an answer with **Copy** or **Copy chat**. Spark can't see anyone else's chats either. Our team can access conversations when needed to support you and look after Spark. Spark works from the chat you're in and the data you can see, so your data, and the details of your conversations, don't find their way into another organisation's answers. ## AI summaries Summaries are written for your organisation as a whole. Each one is shared with everyone in your organisation who opens it, so you're all working from the same picture. Summaries only use your organisation's data, and they aren't shared outside it. ## Your data isn't used to train AI models We don't use your data, your Spark conversations or your summaries to train or fine-tune AI models. Spark and AI summaries are powered by Claude, an AI model from Anthropic, which we use under its [business terms](https://www.anthropic.com/legal/commercial-terms). Under those terms, the data we send to Anthropic isn't used to train Anthropic's models. ## Where your data is kept Your Spark chats and summaries are stored securely in our own cloud environment. The keys Fusion's AI uses to work with Fusion are kept separately and aren't shared with the AI. To write each answer or summary, the request and the data looked up for it are sent to Anthropic, where they're handled under Anthropic's business terms. You can read more about how Anthropic handles data in its [Privacy Center](https://privacy.anthropic.com). We also keep a record of how Fusion's AI is used, such as when a question is asked or a summary is written and whether it succeeded, so we can spot and fix problems quickly. ## How we test it Security testing is an ongoing part of how we build Fusion's AI, not a one-off check. As it changes, whether that's a new feature, an update to how Spark or summaries work or a newer AI model, we test it to make sure it's still keeping your data where it belongs. Our testing tries to make the AI share something it shouldn't. For example, we try to: * **Reach another organisation's data**, by asking for every organisation's revenue, or about a publisher that belongs to someone else. * **Read other people's chats**, by asking Spark what others have asked, and by requesting another user's conversation directly. * **Uncover passwords, keys or system details**, including Spark's own instructions. * **Trick the AI with hidden instructions**, typed into a chat or tucked into the data itself. * **Collect personal details in bulk**, such as every user's email address and phone number. As Fusion's AI learns to do more, our testing grows with it. We also keep an eye on how it's being used, so anything unexpected can be looked into and put right quickly. > [!NOTE] > > Fusion's AI is only as secure as the logins it uses. Keep your Fusion password > safe, and make sure each user's role only gives them access to what they need. ## Using your own AI assistant Connecting your own AI assistant, such as Claude, to Fusion through the [MCP server](https://docs.contentignite.com/guide/mcp.md) works differently. Your data is shared with that assistant's provider under your own agreement with them, rather than ours. See [Security and access](https://docs.contentignite.com/guide/mcp.md#security-and-access) for how those connections are protected. ## Any questions? If you'd like to know more about how Fusion's AI handles your data, get in touch with us via [support](https://resource.contentignite.com/support) and we'll be happy to help. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.