AI agents connected to your content through integrations like ChatGPT can read, summarize, and reproduce sensitive files the moment they have access. If an agent can preview a file, it can comprehend and reproduce what is inside. Most enterprises do not realize how much reach these integrations have until they see it demonstrated. This demo shows exactly that, and how Box Shield puts you back in control. In the first half, we connect ChatGPT to Box with the Box plugin enabled and ask it a simple question: what files do you have access to? The answer reveals that ChatGPT can see every folder across every classification level, Public, Confidential, and Internal Only, with no guardrails in place. We then ask it to summarize a file from the Internal Policies folder. It does, instantly, pulling the full content directly from the file. In the second half, we flip to the admin side and configure classification-based access policies in Box Shield. Two policies, two very different rules. For Confidential content, all integrations are allowed except ChatGPT. For Internal Only content, all integrations are blocked except Box on Salesforce. Two classifications. Two distinct access profiles. One control. Back in ChatGPT, we run the same prompts. This time it cannot read the file, cannot preview it, and cannot pull the raw content. Both workarounds fail. The integration restriction holds. That is classification-based access policy in Box Shield: one control tied to your existing content classifications that governs what AI agents and integrations can actually read, not just what they can download. Content stays useful for your teams and safe from the agents you have not approved. FAQs: Q: What is classification-based access policy in Box Shield? A: Classification-based access policy is a security control in Box Shield that governs what AI agents and integrations can read based on how content is classified. Admins can allow or block specific integrations from previewing, viewing, searching, or reading file content, not just downloading it, based on the file's classification label. Q: What is the difference between download restriction and read restriction? A: Download restriction prevents an integration from downloading a file. Read restriction goes further, blocking the integration from previewing, viewing, searching, or reading the file's text content at all. This demo shows why read restriction matters: ChatGPT can summarize a file it can preview even if it cannot download it. Q: Can I set different rules for different content classifications? A: Yes. This demo shows two different policies in action. Confidential content blocks ChatGPT specifically while allowing all other integrations. Internal Only content blocks all integrations except Box on Salesforce. Each classification can have its own distinct access profile. Q: Does this work for third party AI agents beyond ChatGPT? A: Yes. The integration restriction control in Box Shield applies to any integration connected to Box, not just ChatGPT. Admins can allow or block any specific integration by name across any classification level. Q: Do I need to change my existing content classifications to use this feature? A: No. The control is tied to your existing Box Shield classifications. If you already have content classified as Confidential or Internal Only, you can immediately apply integration restrictions to those classifications without reclassifying any content. Q: What happens when ChatGPT tries to work around the restriction? A: In this demo, ChatGPT offers two workarounds when it cannot read the file directly: opening the file in Box preview or pulling the raw file. Both fail. The integration restriction blocks preview access and raw read operations, so neither workaround succeeds.

 165384      7