This chapter deals with applications where watermarking is a security primitive included in a larger system protecting the value of multimedia content. In this context, there might exist dishonest users, in the sequel so-called attackers, willing to read/overwrite hidden messages or simply to remove the watermark signal. The goal of this section is to play the role of the attacker. We analyze means to deduce information about the watermarking technique that will later ease the forgery of attacked copies. This chapter first proposes a topology of the threats in Section 6.1, introducing three different concepts: robustness, worst-case attacks, and security. Previous chapter has already discussed watermark robustness. We focus on worst-case attacks in Section 6.2, on the way to measure watermarking security in Section 6.3, and on the classical tools to break a watermarking scheme in Section 6.4. This tour of watermarking security concludes by a summary of what we know and still do not know about it (Section 6.5) and a review of oracle attacks (Section 6.6). Last, Section 6.7 deals with protocol attacks, a notion which underlines the illusion of security that a watermarking primitive might bring when not properly used in some applications.