In this chapter, a comprehensive investigation of state-of-the- art machine learning (ML)-based spectrum sharing (SS) solutions is presented. While the ML-based SS networks can help improve the spectral efficiency, they can also be a double-edged sword. The dynamic access frameworks introduce more security and privacy risks into the system. When secondary users observe the activity of primary users (PUs), the sensing procedure can be disturbed by the malicious attackers by launching the PUE attacks or SSDF attacks. The attackers may also exploit these opportunities to harm the privacy of PUs. Besides these, the system also suffers the same security issues found in traditional wireless communications, such as jamming attacks and eavesdropping attacks. The chapter discusses these physical layer attacks and potential countermeasures. In the chapter, the latest research in ML-related security is comprehensively surveyed.