The Cyber threats keep getting smarter and happening more often, which makes it very hard for businesses to protect their networks and systems. Traditional defence methods like rule-based and signature-based often can’t keep up with how quickly and easily cyberattacks change and adapt. This research examines how reinforcement learning, a subset of machine learning, might be applied to make cyber security systems more flexible and self-sufficient. Through interacting with their surroundings, reinforcement learning agents can learn the best ways to defend themselves by using the power of trial-and-error learning. This lets them adapt quickly to new risks. This paper proposes a way to use reinforcement learning in cyber defence systems and discuss about important things to think about, like action and state spaces, reward functions, and choosing the right method. The outcomes of experiments show that reinforcement learning can make cyber defence more effective and efficient compared to standard methods. This study lays the groundwork for making cyber defence systems smarter and more flexible so they can deal with cyber threats that are always changing.