Quantum key distribution's (QKD's) central and unique claim is information theoretic security. However there is an increasing understanding that the security of a QKD system relies not only on theoretical security proofs, but also on how closely the physical system matches the theoretical models and prevents attacks due to discrepancies. These side channel or hacking attacks exploit physical devices which do not necessarily behave precisely as the theory expects. As such there is a need for QKD systems to be demonstrated to provide security both in the theoretical and physical implementation. We report here a QKD system designed with this goal in mind, providing a more resilient target against possible hacking attacks including Trojan horse, detector blinding, phase randomisation and photon number splitting attacks. The QKD system was installed into a 45 km link of a metropolitan telecom network for a 2.5 month period, during which time the system operated continuously and distributed 1.33 Tbits of secure key data with a stable secure key rate over 200 kbit/s. In addition security is demonstrated against coherent attacks that are more general than the collective class of attacks usually considered.