liveBook · Manning

Purpose lock

As explained in AI Governance

A purpose lock is a design control that encodes strict boundaries on the purposes for which an agent may process data and use tools.

An agent can expose internal data through outbound tool calls. Dynamic MCP tool discovery can also add unvetted tools without notification, while each call may create a new data-processing relationship. A purpose lock encodes strict limits on the purposes for which the agent may process data and use tools. section 5.6.1

Enforce the limits with a purpose-aware egress gate that checks every outbound tool call before data leaves the system. OPA can decide whether a call is allowed, but does not intercept or block it; a separate enforcement point is required. Configure the gate to fail closed when no rule permits a call. section 5.6.1

Before permitting a tool call, check:

  • Whether the proposed use matches the purpose for which the data was collected. section 5.6.1
  • Whether the data fields sent are necessary for that purpose. section 5.6.1
  • Whether the destination tool or region is permitted. section 5.6.1

How AI Governance covers it

How deep it goes
Operationalizes purpose, necessity, and destination checks through a fail-closed gate, while distinguishing policy decisions from enforcement.
Examples
Frames a policy decision as whether an agent may send specified fields to a specified tool.
What you can do after
Apply purpose, necessity, and destination checks at each outbound call, using OPA for decisions and a separate enforcement point.
How it connects
Connects purpose limitation and data minimization with GDPR, MCP tool discovery, OPA, and outbound tool-call enforcement.

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.