Modern processors routinely execute instructions that never retire, but their effects are not necessarily invisible. In this talk, we explore how these transient instructions can leave observable traces in the microarchitecture and how attackers exploit them to break security guarantees. Starting with cache side channels, we examine how speculative execution can be exploited through branch predictor mistraining, leading to Spectre attacks. We then investigate Meltdown-type vulnerabilities, where faulting instructions can transiently expose sensitive data, and discuss further attacks involving internal CPU buffers and registers. We examine the underlying microarchitectural causes of these vulnerabilities and discuss the effectiveness and limitations of existing mitigations. The talk illustrates how performance optimizations in modern processors can undermine security boundaries, even when architectural correctness is preserved.