A technology is described for determining a trust score or trust metric used to determine whether to allow a connection to a high risk destination. In one example of the technology, a connection request to initiate a connection with a destination over a network may be received at a communication service. The communication service may identify a customer account associated with the connection request and obtain a trust score that indicates whether the connection request associated with the customer account may be allowed. The communication service may determine whether the connection request associated with the customer account is allowed to initiate the connection with the destination based on the trust score, and initiate the connection with the destination over the network when the trust score allows the connection request.