Abstract: A safety event is determined as affecting a user based at least in part context data collected at a user device associated with the user. In some aspects, context data is detected from sensors on the client device, the context data describing a present context of the user. A deviation of the present context from a historical context is determined to be beyond a threshold. Determining that the deviation is beyond the threshold can be determined to correspond to a safety event potentially jeopardizing safety of the user. In some aspects, an action can be launched in response to determining the safety event.

Abstract: Code of a particular application is analyzed against a semantic model of a software development kit of a particular platform. The semantic model associates a plurality of application behaviors with respective application programming interface (API) calls of the particular platform. A set of behaviors of the particular application is identified based on the analysis of the code and a particular one of the set of behaviors is identified as an undesired behavior. The particular application can be automatically modified to remediate the undesired behavior. The particular application can be assigned to one of a plurality of device modes, and access to the particular application on a user device can be based on which of the plurality of device modes is active on the user device.