Every year, millions of letters/parcels containing illicit goods are detected by customs authorities, which use traditional security screening equipment. However this equipment cannot detect all kinds of illicit goods and the detection procedure heavily depends on the attention of the customs officer. In order to achieve sufficiently fast intelligent screening of the large volumes of letters/parcels and detect all common kinds of threats, this paper proposes a highly innovative architecture well-beyond the state-of–art. In particular the proposed architecture monitors every letter/parcel by incorporating: (a) terahertz/X-ray sensors, (b) chemical, biological, radiological and nuclear (CBNR) sensors, (c) artificial robot-noses for narcotics, explosives etc., (d) magnetometers for weapons, firearms, banknotes etc., (e) acoustic sensors for liquids/gases/solids, (f) weight/pressure sensors to measure weight distribution, size and shape. Sensory information can be: (a) used to create a “Spectral Signatures Dictionary of Illicit Goods and Threats”, (b) fused to segment/isolate illicit goods and (c) visualized in the form of annotated high-resolution tensor-structured (3D/4D) multisensory image data. The proposed solution also gathers available information for the sender/recipient from various resources, while it also analyzes data from the dark web. All information is forwarded to an AI-based knowledge infrastructure.