> If ChatGPT is able to emit output that is watermarked

Then the watermark should be made public. We need a way to tell ai-generated content from human-generated content without actually having to read it.