Autonomous vehicles (AVs) will radically change the transportation landscape in the coming years. AVs' localization and navigation systems primarily depend on the global navigation satellite systems (GNSS). However, the lack of encryption and low signal strength make GNSS signals vulnerable to intentional and unintentional threats. Out of all intentional threats, spoofing is a sophisticated and detrimental type of attack in which an attacker can manipulate and override the original GNSS signal, and an AV's GNSS receiver receives false position and tracking information – misled by the spoofer. In this way, an AV can be incrementally directed to the wrong destination, compromising the user's safety and security. This paper presents a novel approach for detecting GNSS spoofing attacks where clock correction polynomial parameters, ephemeris parameters, and integrity data (CEI) from the visible satellites are used to detect a spoofing attack in real time. We hypothesize that the pattern of certain CEI time series variables could change during a spoofing attack. Therefore, understanding and learning the pattern of CEI variables will help identify anomalies in GNSS signals at any given instance. Our approach investigates a recurrent neural network, a Long Short Term Memory (LSTM) network, trained with authenticated GNSS signals over a duration to detect the anomaly in the variable values due to an attack. GNSS datasets may include satellite signal data from different satellite systems, such as GPS, Galileo, GLONASS, and BeiDou, which are used to train the AI model. The AI-based detection model is evaluated using the GNSS spoofing attack and attack-free datasets. The evaluation results prove the potential of our approach toward a robust solution.