Ransomware has emerged as one of the most pervasive threats in modern cybersecurity, causing significant damage through the encryption of critical data and subsequent demands for ransom payments. A novel universal network-level filter has been introduced to address the limitations of traditional detection systems, which often fail to adapt to the evolving techniques used by adversarial attacks. Through the integration of machine learning models and real-time network traffic analysis, the proposed filter dynamically identifies and mitigates both known and novel ransomware threats with minimal latency. The system's ability to generalize across diverse network environments, coupled with its scalability and adaptability, positions it as a critical advancement in ransomware defense. Experimental results demonstrate the filter’s effectiveness in reducing false negatives and maintaining high detection accuracy, even under adversarial conditions and high-traffic scenarios, highlighting its robustness and practical application in enhancing network security infrastructures.