liveBook · Manning

Tool misuse and exploitation

As explained in AI Governance

Tool misuse is an attack in which an agent is manipulated into improperly using access it already has.

Tool misuse occurs when an attacker manipulates an agent into improperly using access it already has. Unlike an unauthorized transfer blocked by permissions, misuse can exploit approved actions in patterns that look individually normal; monitoring behavior matters alongside enforcing permission boundaries. section 4.8.1

Small refunds, fraudulent pattern

An attacker manipulates ClaimAssist, which has access to a payment API, into issuing many small refunds to legitimate accounts. Each refund may look normal, while the overall pattern is fraudulent. section 4.8.1

How AI Governance covers it

How deep it goes
Defines the attack and contrasts direct unauthorized transfers with subtler fraudulent patterns, but does not detail a full defense procedure.
Examples
Uses ClaimAssist payment access and repeated small refunds to legitimate accounts as a fraud scenario.
What you can do after
Understand how an attacker can manipulate an agent into abusing approved access and why behavior patterns matter.
How it connects
Connects tool misuse to agents, API access, permission systems, behavioral monitoring, and other agent-specific risks.

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.