Conventional authentication systems, relying on one-factor or two-factor methods like passwords and hardware tokens, have historically shown significant vulnerabilities. To overcome these shortcomings, we introduce an innovative continuous authentication framework that leverages a user’s distinct linguistic and behavioral fingerprint, derived from routine device interactions. Our approach captures a range of features—such as typing speed, spelling error tendencies, slang, acronyms, and vocabulary preferences—to profile the user’s writing style across diverse contexts. Unlike traditional keystroke dynamics or touch-based biometrics, this method adapts to shifts in tone or language, accommodating scenarios from formal correspondence to casual, multilingual exchanges. We employ a BiLSTM-based model, trained exclusively on user data to create robust, personalized profiles. Our research emphasizes privacy-preserving on-device processing and compares multiple anomaly detection metrics using the BiLSTM model, selecting the most effective methods to ensure reliable authentication. Real-time anomaly detection occurs locally, flagging potential impersonation based on detected deviations during typical device use.