Google already admitted they are fingerprinting generative video and have a safety obsession so I guarantee they do it to their LLMs. Another reason is to pollute the output that folks like Deepseek are using to train derivative models.