As global agriculture shifted toward automated, IoT-driven hydroponic systems to ensure climate-resilient food sustainability, the convergence of digital control and biological processes introduced a critical vulnerability: Cyber-Biosecurity. In smart greenhouse environments, the regulation of nutrient delivery, pH levels, and atmospheric conditions relied entirely on high-velocity sensor data. However, these Cyber-Physical Systems (CPS) were found to be increasingly susceptible to False Data Injection (FDI) attacks. Malicious actors were able to subtly manipulate sensor readings to induce plant stress or total crop failure while remaining undetected by traditional threshold-based alarm systems. This research developed a robust, AI-driven anomaly detection framework designed for the unique temporal demands of agricultural IoT. Central to this framework was the application of Temporal Convolutional Networks (TCN). Unlike traditional Recurrent Neural Networks (RNNs) or LSTMs, TCNs utilized dilated causal convolutions and residual blocks to capture long-range temporal dependencies with significantly lower computational overhead and improved gradient stability. This rendered the model ideal for edge deployment in greenhouse controllers where real-time inference was vital. The study demonstrated how a TCN-based architecture successfully identified sophisticated, low-magnitude data deviations that signaled an ongoing cyber-attack.  This work explored the transition from general network security to a specialized cyber-biosecurity approach by treating the smart greenhouse as a piece of critical national infrastructure. The findings provided a scalable roadmap for securing the automated systems of modern agriculture against digital interference, ensuring that food production remained both resilient and invulnerable.