Utarbeidet med KI.
What risks are associated with an AI agent?
An agent on the website speaks to customers on behalf of the company. It is worth considering what could go wrong before it is published.
The greatest risk is not that the agent says, “I don’t know,” but that it gives an incorrect answer convincingly. The remedy is to limit the sources it uses to answer and to route the most important questions through defined topics rather than allowing free-form responses.
If the agent is directed to internal documents, it may reproduce something that was never intended for customers. Deliberately limit the source, and test it with questions designed to coax out internal information.
Customers enter national identity numbers and account numbers without being asked to do so. Clarify where the conversations are stored, for how long, and who has access — before the agent is deployed, not afterwards.
Set aside time to try to get the agent to do something stupid before customers do. It’s an inexpensive hour.
Still et spørsmål eller del hva som hjalp deg.
Del et spørsmål eller en refleksjon.
Bli den første som bidrar.