CensorX: AI powered real-time content monitoring and safety app
No Thumbnail Available
Date
2026-06
Journal Title
Journal ISSN
Volume Title
Publisher
AIKTC
Abstract
The rapid growth of digital communication platforms has significantly increased users’
exposure to online harassment, hate speech, and explicit content, particularly within private and
encrypted messaging environments. While these platforms provide enhanced privacy through
end-to-end encryption, they also limit the effectiveness of traditional moderation systems.
Existing solutions primarily rely on server-side or cloud-based content analysis, which
introduces concerns related to user privacy, higher latency, and limited applicability in
encrypted communication systems.To address these challenges, this paper proposes CensorX,
an on-device, real-time harassment detection and content censoring framework designed
specifically for Android-based systems. Unlike conventional approaches, the proposed system
operates entirely on the client device, ensuring that sensitive user data is not transmitted to
external servers. CensorX leverages lightweight Natural Language Processing (NLP) models
combined with rule-based filtering techniques to analyze textual content in real time. The
system utilizes Android system-level services to monitor user input across applications without
requiring modifications to third-party platforms. The core methodology involves a hybrid
detection mechanism where a pre-trained NLP classification model identifies harmful or
abusive patterns based on contextual understanding, while a rule-based layer enhances detection
accuracy by capturing explicit keywords and predefined patterns. This combination enables
efficient processing with low computational overhead, making it suitable for continuous
background operation on mobile devices. Experimental evaluation of the proposed system
demonstrates promising performance. CensorX achieves a detection accuracy of 92.4%, with a
precision of 90.1% and recall of 93.6% in identifying harassment-related content. The system
maintains an average processing latency of less than 120 milliseconds per input, ensuring realtime
responsiveness. Additionally, CPU usage remains below 8% on average, indicating
minimal performance impact on the device. These results highlight the effectiveness and
feasibility of deploying on-device AI-based moderation systems for enhancing digital safety
while preserving user privacy.
Keywords: Harassment Detection, Content Censoring, On Device AI, Natural Language
Processing, Mobile Security, Digital Safety, Android Accessibility .