How do you see traditional anomaly detection approaches adapting as attackers increasingly randomize or blend their behavioral fingerprints to resemble normal social media activity? Could integrating contextual or graph-based techniques improve anomaly detection resilience against such adaptive evasion tactics?

Your work addresses how detection systems rely on consistent fingerprint-based signals to differentiate benign vs malicious actors. With attackers increasingly using benign-looking fingerprint profiles (device + behaviour + network) on social media, what metrics or signals (beyond simple fingerprinting) did you identify as resilient enough to still detect attacks? How do you see these evolving in...