The evolution of software systems into more complex ecosystems creates new challenges in ensuring their safe and secure behavior. As the complexity of software ecosystems is inherently higher than regular systems, existing safety mechanisms are no longer reliable in their context. This paper introduces a research path towards adaptive safety mechanisms that can support the degree of dynamism and high level of uncertainty introduced by these systems of systems. Our planned approach is to use runtime trust evaluation as a decision factor when enabling or disabling safety features on demand.