AuthAi: Real-Time Behavioral AI for CAPTCHA-Free Security
Piyush Raj
BE-CSE, Chandigarh University 1201piyushraj@gmail.com
Priyanshu Kumar
BE-CSE, Chandigarh University priyanshu345kumar@gmail.com
Nikhil Kumar Tiwari
BE-CSE, Chandigarh University nikhiltiwariofficial2608@gmail.com
Divyanshu Dwivedi
BE-CSE, Chandigarh University divyanshudwivedi161@gmail.com
Sandeep Kaur
CSE, Chandigarh University sandeep.e3095@cumail.in
Abstract—AuthAI is a marked improvement over traditional, more static methods of authentication (like CAPTCHA - which many may agree is more of a hindrance than a measure of assurance) that typically try to use ”prove you’re not a robot” rituals. Instead, AuthAI uses a continuous behavioral authentica- tion strategy. To help implement this system, researchers designed a complex combination of models, where Transformers provide the most accurate scoring, and XGBoost and Autoencoders to support anomaly detection and provide resilience. Built-in adversarial simulations can help guard against mimicry attempts, and adaptive feedback will allow the framework to learn from misclassifications, and further refine its defense over time.There is an analytics dashboard, which also improves transparency by informing the user Key Features and Confidence levels - something useful for researchers and practitioners alike! The technology is still very limited, at this point, in terms of the relatively small and narrow dataset used to train it. But, the team already has plans for the future - richer, multimodal signals; the data set size increased; federated learning; the inclusion of large language models to increase explainability.
Index Terms—Authentication, Bot Detection, Behavioral Bio- metrics, Keystroke Dynamics, Mouse Movements, Click Rates, Navigation Patterns, Error Frequency, Multi-Model Ensem- ble, Transformer, LSTM, Autoencoder, Random Forest, XG- Boost, Isolation Forest, Adversarial Simulation, Adaptive Feed- back Loops, Anomaly Detection, Explainability, Analytics Dash- board, Scalability, Robustness, Continuous Authentication, User- Friendly, Federated Learning, Multimodal Signals.
