The use cases listed in Annex III that concern the use of an AI system by public authorities consistently use the terms ‘on behalf of’ or ‘or on [their] behalf’to cover AI systems that are intended for use not only by the public authorities themselves, but also by third parties (i.e. private entities) where a public authority outsource s activities to those entities falling within th ose use cases.
The use of those terms is meant to avoid circumvention of the high-risk classification in cases where an AI system is marketed only to certain natural or legal persons, while the risk of the system for fundamental rights is the same as that where public authorities are involved, because the activity is performed on behalf of the public authority.
A third-party/private entity will be considered to act ‘on behalf of’ a public authority if the public authority delegates the performance of certain activities (or parts of those activities) to th at party/entity or that authority has requested the party/entity to support such activities in specific cases. A third-party/private entity will not be considered to act ‘on behalf of’ a public authority if the private entity acts on its own behalf, either on a voluntary basis or to comply with a legal obligation.
Example: An accounting firm that deploys an AI system that detects money laundering to comply with its obligations under EU anti-money laundering legislation will not be considered to act on behalf of law enforcement authorities, but on its own behalf, so that the system will not be classified as high-risk in the area of law enforcement listed in point 6 of Annex III AI Act.