liveBook · Manning

Data oversharing

As explained in AI Governance

Oversharing is the exposure of information to a broader audience than intended through overly permissive sharing or access settings.

Data oversharing exposes information to a broader audience than intended. GenAI can make overly broad permissions more dangerous: even when a system respects access controls, semantic retrieval can surface relevant files that users would not have found through keyword search. section 4.6.2

Enterprise AI exposes latent permission problems

Microsoft 365 Copilot surfaced payroll, legal, and strategy files that employees could access but were not meant to see. Research cited in the discussion found 15% of business-critical files overshared and more than 3% of sensitive business data shared organization-wide. Concerns about oversharing delayed many Copilot rollouts. A separate Salesforce Agentforce case involved a domain that remained allowlisted after expiring; researchers bought it for $5 and regained control of that destination. section 4.6.2

How AI Governance covers it

How deep it goes
Explains permission and retrieval risks through enterprise cases but stops short of prescribing remediation.
Examples
Uses Microsoft 365 Copilot file exposure and a Salesforce Agentforce expired-domain case.
What you can do after
Evaluate how broad permissions and semantic retrieval can expose information users were not meant to see.
How it connects
Connects data oversharing to access controls, semantic search, and content security policies.

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.