Systems that support mobile agents are increasingly being used on the global Internet. Security concerns dealing with the protection of the execution environment from malicious agents are extensively being tackled. We concentrate on the reverse problem, namely how a mobile agent can be protected from malicious behaviour of the execution environment, which is largely ignored. We will identify the problem of trust as the major issue in this context and describe a trusted and tamper-proof hardware that can be used to divide this problem among several principals, each of which has to be trusted with a special task. We show that the presented approach can be used to mitigate an important problem in the design of open systems.