liveBook · Manning

Confused deputy problem

As explained in AI Governance

The confused deputy problem is a security flaw in which a program with legitimate authority is tricked into using that authority on behalf of someone who lacks it.

A confused deputy is a program tricked into misusing its legitimate authority. In an AI assistant, a broadly privileged service account can let users retrieve information they could not access themselves. section 4.6.1

Reduce the risk by limiting what the assistant can access and whose permissions it uses:

  • Query data using the requesting user's identity, passing that identity through the assistant and connected data components. section 4.6.1
  • Exclude sensitive data the assistant does not need, reducing the impact of a future compromise. section 4.9

ClaimAssist

When ClaimAssist connected to customer records, it queried data as the requesting customer and excluded fraud investigation, personnel, and financial account records. Customers could receive their own claim information, not other customers' records. section 4.9

How AI Governance covers it

How deep it goes
Defines the flaw and explains user-context queries, including the need to propagate identity through connected components.
Examples
Contrasts a broadly privileged chatbot involved in a data breach with ClaimAssist's customer-scoped queries and excluded records.
What you can do after
Configure assistant queries to use the requesting user's identity and exclude unnecessary sensitive data.
How it connects
Links authorization, service accounts, and identity propagation in enterprise data access.

Also explained in

Compare how 2 books explain confused deputy problem →

Related concepts

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.