we investigate the functional limitations, environmental costs, and security vulnerabilities inherent in modern artificial intelligence and the Transformer architecture. Research from MIT and various technical papers highlights how AI faces "model collapse" when trained on synthetic data, as well as "catastrophic forgetting" where new information causes the system to lose prior knowledge. Mathematical analyses demonstrate that Transformers struggle with function composition and complex logic, often leading to factual hallucinations and reasoning errors. Furthermore, the texts identify prompt injection attacks as a significant security risk, where malicious instructions can bypass safety guardrails to leak data or spread misinformation. Collectively, the documents suggest that while AI is transformative, it remains constrained by technical bottlenecks, reliability issues, and high resource consumption. Efforts toward achieving Artificial General Intelligence must therefore overcome these fundamental obstacles through better data quality and enhanced architectural robustness.