liveBook · Manning

Contextual security research framework

As explained in AI Governance

A contextual security research framework defines which actions are in scope for a task and checks proposed actions against task-specific rules.

A contextual security framework limits an agent to actions appropriate for its specific task. Before work begins, it generates a task-specific policy; an enforcer checks each proposed action against that policy before execution. section 8.3

Separate task scope from approval:

  • Define which actions are in scope for the task. section 8.3
  • Check proposed actions against the policy before they reach an outside system; retain human approval for sensitive actions when needed. section 8.3

Urgent work email

A policy might allow replies to approved work addresses while blocking message deletion and emails to outside recipients. section 8.3

How AI Governance covers it

How deep it goes
Details policy generation before a task and enforcement before each proposed action runs.
Examples
Uses urgent email replies to distinguish approved recipients from blocked deletion or external sending.
What you can do after
Apply it by generating a task-specific action policy and checking proposals before execution.
How it connects
Pairs with human approval and relates to policy enforcement and agent governance.

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.