In today's digital world, propaganda and disinformation spread like wildfire through various channels, including social media. Researchers are constantly working on innovative ways to detect and combat these tactics, and Arabic content presents unique challenges. The recent ArAIEval Shared Task at the ArabicNLP 2024 conference focused on this crucial area. This task challenged participants to create AI models to identify propaganda techniques hidden within Arabic text and multimodal memes. One of the core challenges was identifying specific propaganda techniques like "Name Calling," "Appeals to Fear," or "Straw Man" within tweets and news articles, even pinpointing the exact words used. For memes, the task was to discern whether the combination of image and text conveyed a propagandistic message. A key finding was that fine-tuning pre-trained language models like AraBERT proved to be a highly effective approach for this task. Several teams employed creative methods like data augmentation (expanding the training data with similar examples) and clever ways to encode and decode labels to improve the model's accuracy. But it wasn't without its difficulties. One hurdle was the imbalanced nature of the datasets. Some propaganda techniques appear far more frequently than others, making it harder for models to learn the less common ones. Another area for improvement lies in addressing the inherent biases that can creep into datasets due to the subjective nature of propaganda and the diversity of Arabic dialects. Overall, despite these limitations, ArAIEval demonstrated how AI can be a powerful tool for unmasking propaganda techniques in Arabic content. Future research will address the challenges and explore additional complexities in this ongoing battle against misinformation.