Reconfigurable Intelligent Surfaces (RISs) are expected to be massively deployed in future Beyond-5th generation (B5G) wireless networks, thanks to their ability to programmatically alter the propagation environment, inherent low-cost and low-maintenance nature. Indeed, they are envisioned to be implemented on the facades of buildings or on moving objects. However, such an innovative characteristic may potentially turn into an involuntary negative behavior that needs to be addressed: an undesired signal scattering. In particular, RIS elements may be prone to experience failures due to lack of proper maintenance or external environmental factors. While the resulting Signal-to-Noise-Ratio (SNR) at the intended User Equipment (UE) may not be significantly degraded, we demonstrate the potential risks in terms of unwanted spreading of the transmit signal to non-intended UEs. In this regard, we consider the problem of mitigating the unwanted effect caused by faulty RIS elements by proposing two simple yet effective algorithms, which are based on maximizing the Signal-to-Leakage-and-Noise-Ratio (SLNR) over a predefined two-dimensional area and are applicable in the case of perfect channel-state-information (CSI) and partial CSI, respectively. Numerical and full-wave simulations demonstrate the added gains compared to leakage-unaware and reference schemes