In response to the emerging threat of AI-enabled penetration attacks, characterized by their high level of automation and adaptability, this paper proposes a multi-stage proactive defense strategy. Initially, the strategy is creating mechanisms for security situational awareness. Followed by the dynamic deployment of honeypots to disrupt the flow of information to intelligent attacks, the strategy culminates in the implementation of a dynamic moving target defense (MTD) mechanism, based on security situational values, to further refine the techniques of information camouflage and confusion. This multifaceted strategy effectively stalls the adaptive mechanisms of sophisticated attacks. The strategy’s validation was conducted on the NASim platform, with the simulation results demonstrating its effectiveness in curtailing intelligent attacks, particularly those powered by reinforcement learning. This strategy proved more effective than relying solely on honeypots or MTD mechanisms, with a notable increase in the average attack steps from 57.11 to 435.65, compared to scenarios without any defense measures.