Microsoft Purview: Analysts can deanonymize AI prompts and responses

Microsoft's Purview is a management and analysis tool that allows IT teams to monitor AI prompts and responses, for example, using “Insider Risk Management.” By default, users are pseudonymized; they cannot be directly identified or assigned. Microsoft is now providing analysts in Purview with the ability to deanonymize users despite active anonymization, “while preserving privacy protections.” This is reminiscent of a square circle.

Microsoft has specified this in the Microsoft 365 Admin Center in announcement MC1304292 (copy at merill.net). Purview Insider Risk Management gains the ability for analysts to gain insight into users' AI prompts and responses. This is possible even with anonymization enabled. This is intended to increase visibility of AI-related risks while ensuring role-based access controls. Insider Risk Management in Purview aims to detect risks such as data leaks, intellectual property theft, or security policy violations. Users are pseudonymized by default, and analysts are only allowed to access data according to their roles, which is intended to protect privacy.

Rollout from early May 2026

Microsoft will begin rolling out the public preview from early May, with completion expected by mid-month. General availability will follow from mid-June. Regarding who is affected, Microsoft states that only admins and analysts using Purview Insider Risk Management, as well as organizations investigating AI-related insider risks, are affected by the change – users, on the other hand, are not considered affected by Microsoft.

According to Microsoft, the specific changes include that analysts can now view AI interaction messages, including prompts and AI-generated responses, provided they are associated with insider risk indicators. The interactions are visible even when anonymization is enabled. User identities remain pseudonymized until authorized analysts perform a deanonymization.

Existing role-based access controls, audit logging, and privacy safeguards remain in place. Microsoft intends to enable the new feature by default. Therefore, admins do not need to do anything, Microsoft writes.

Microsoft's relationship with privacy appears to be rather loose. Last week, the company announced that it intends to take stronger account of the confidentiality labels assigned in the automated analysis of files in the future.

(dmk)