An organization wants to prevent employees from using unauthorized consumer AI tools (like public ChatGPT) on corporate devices. They want to block uploads of sensitive files to these services while allowing approved enterprise AI tools. Which Microsoft solution enforces this endpoint-level control?

A is correct because Microsoft Purview DLP endpoint policies (deployed via Microsoft Defender for Endpoint) can detect when a user attempts to upload a file containing sensitive content or bearing a sensitivity label to cloud services including web browsers. An 'endpoint activities' DLP rule can list unauthorized AI services (e.g., chat.openai.com) as restricted domains and block file uploads to them, while allowing uploads to approved enterprise AI services. This provides granular control at the file-sensitivity level. B is wrong because Azure Firewall blocks traffic at the network perimeter, not on individual endpoints; it would block all traffic to a domain including legitimate personal use outside corporate time, and does not distinguish between uploading a sensitive labeled file versus general browsing. C is wrong because Conditional Access blocks application sign-in for corporate identities; consumer AI tools often do not require corporate identity sign-in, so Conditional Access is ineffective for blocking them. D is wrong because Defender for Endpoint web content filtering blocks entire site categories (like social media or adult content) at the network level; it cannot distinguish between uploading sensitive files versus normal browsing on an AI site.