Deployments of AI solutions in organizations are no longer seen merely as experiments and are becoming a standard element of Microsoft 365 environments. After enabling AI capabilities, however, practitioners have observed new vectors of data exposure often linked to improper sharing settings and permissions.
Sharing controls in a cloud environment determine which content is accessible to users and systems. Incorrectly configured permissions or overly broad sharing rules can make documents or data available to models and services that should not process or reveal them. This is not primarily a flaw of the AI technology, but a consequence of insufficient control over resource access.
AI models operating within a corporate environment may use accessible content and reference it in responses. When resources are overly shared, there is a risk that AI outputs will include fragments of sensitive information or summaries of data that should remain protected. Deploying AI therefore changes the risk profile related to access management.
Organizations should promptly review three areas: user and group permissions, sharing policies across the environment, and AI integration configurations. The audit should examine who has access to which resources, what links and sharing rules are active, and how system settings route content to AI processing.
Based on audit findings, implement access restriction by necessity, tighten external sharing rules, and verify AI integration settings to exclude processing of sensitive materials. Monitoring activity and reviewing logs after deployment, as well as training users on secure sharing practices, are equally important.
Implementing AI in Microsoft 365 requires parallel control over sharing and permissions. To reduce the risk of unintended data exposure, organizations must urgently audit permissions, sharing policies and AI configurations and apply concrete controls and monitoring.
Source: https://itwiz.pl/ai-pokazuje-wiecej-niz-powinno-problem-lezy-w-udostepnieniach-w-microsoft-365/