In ambient systems, a principal may be a physical object whose identity does not convey useful information for taking security decisions. Thus, establishing a trusted channel with a device depends more on the device being able to demonstrate what it does, rather than who it is. This paper proposes a security model that allows a principal to establish the intent of an adversary and to make the adversary prove its trustworthiness by furnishing proof of current and past behavior.Key-words: Security, ambient systems, trust, programming model.