The thinnest path problem (TPP) seeks to transmit a message from one node to another in a wireless network while ensuring the message is exposed to as few nodes as possible. This problem holds significant importance in the domain of wireless network security and privacy, where minimizing overhearing nodes can prevent unauthorized access and eavesdropping. Despite its importance, the thinnest path problem has yet to be well studied. In this paper, we introduce a novel framework based on Reinforcement Learning that leverages Hypergraph Neural Network (HGNN) to learn dynamic features in various states during the construction of the solution. The use of HGNN enables the model to effectively capture complex relationships between network nodes during wireless transmission, leading to more precise and efficient pathfinding. Experimental results indicate that our proposed framework not only outperforms existing learning-based baselines in terms of solution quality in competitive time, but also is significantly more efficient compared to traditional solver. This demonstrates the potential of hypergraph learning in addressing complex combinatorial optimization problems in wireless networks, paving the way for future research in this field.