Standardized near-universal watermarking for AI-generated text creates a defensible line for what counts as deception. If everything is watermarked by default, then watermarked text is okay with no further disclosure necessary, but if it’s AI-generated and not-watermarked then you must have taken steps to remove the watermark, which is clear-cut lying.
Standardized near-universal watermarking for AI-generated text creates a defensible line for what counts as deception. If everything is watermarked by default, then watermarked text is okay with no further disclosure necessary, but if it’s AI-generated and not-watermarked then you must have taken steps to remove the watermark, which is clear-cut lying.