liveBook · Manning

Principle of least privilege

As explained in AI Governance

Least privilege is the security principle of granting a system or user only the minimum access needed to perform its task.

Least privilege means giving an AI system only the data, tools, and permissions its task requires. Broad access can turn an assistant into a confused deputy: it may use its credentials to access or expose information beyond what its user needs. section 4.6.1

Apply the principle as an operating rule:

  • Start with the minimum data and tool access the use case needs; expand only when production evidence supports the change. section 8.4
  • Limit tool use and sensitive data exposure, require approval for selected actions, and record delegated access and its limits. section 8.4
  • Keep records sufficient to reconstruct events, monitor risky behavior, and define when to pause or review the system. section 8.4

Overbroad support-app access

Attackers who obtained credentials with broader permissions than a chatbot needed used them to query and export support records from more than 700 organizations’ Salesforce systems. The incident illustrates how excess access can expose sensitive records and credentials for follow-on attacks. section 4.6.1

How AI Governance covers it

How deep it goes
Operationalizes least access through scoped tools and data, approvals, records, monitoring, and review conditions.
Examples
Uses a chatbot credential incident involving exported Salesforce support records to illustrate excessive permissions.
What you can do after
Configure narrower data and tool access, then expand permissions only when production evidence supports the change.
How it connects
Links least privilege to confused deputy risk and delegated access in AI agents.

Also explained in

Compare how 22 books explain principle of least privilege →

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.