Enterprise software vendors market synthetic data as a legal workaround for privacy regulations and data scarcity, promising that artificially generated rows can bypass GDPR enforcement and feed machine learning pipelines. However, this pitch ignores software engineering realities, as synthetic data is algorithmically constructed to mimic a source dataset's statistical distribution but introduces risks like model collapse, neural memorization leaks, and silent performance degradation. Technical generation mechanisms include rule-based mocking, which uses explicit constraints and regular expressions for infrastructure load tests but lacks statistical intelligence; parametric and Bayesian models, which map joint probability distributions and carry over column correlations from seed data; and deep generative architectures like Generative Adversarial Networks (GANs), where a generator and discriminator train in a minimax game, Variational Autoencoders (VAEs), which map records into a latent space for reconstruction, and transformer architectures adapted from large language models.