Watermarking techniques for LLM-generated code are not robust due to easy removal by semantic-preserving transformations.
