Memorial urging AI watermarking fails amid enforcement and scope concerns

Delegates debated a memorial that would recommend Congress require generative AI systems to mark generated images, audio and text with an "AI" watermark and to police removal. Proponent Camille Bringhurst framed the proposal as a public-safety and creators'-rights measure, saying "AI videos are being posted all over the Internet deceiving people to think that they're real."

Questions from multiple delegates targeted enforceability: how an audio watermark would work, who would bear penalties, and whether older AI models would need retrofitting. Audrey LaPlante asked, "What if the AI becomes sentient and refuses to watermark itself?" — an exchange the proponent answered by pointing to coded constraints and the memorial's nonbinding, federal-recommendation nature.

Opponents and skeptics highlighted practical circumvention (cropping, editing, off-shore models) and the memorial's lack of specified penalties or enforcement mechanisms. Delegates repeatedly noted the memorial applies to companies (not individual users) and that platforms and the federal government would carry enforcement burdens.

The chamber called for a division to count votes; after the count the chair announced the memorial did not secure a majority and did not pass. The transcript records the substance of the technical questions and the proponent's repeated emphasis that the memorial is a recommendation to Congress rather than a detailed regulatory framework.