Fluent comments, survey answers, and essays can now be generated at scale by large language models at almost no cost, with no detectable pattern of duplication. Institutions that rely on public input can no longer assume that each submission reflects a distinct person who formed the view and can defend it. The New York attorney general found that nearly 18 million of 22 million comments on a 2017 net neutrality proposal were fabricated. Earlier fraud left technical traces. Personalized machine-generated text removes those traces while preserving the appearance of genuine participation.
The difficulty is not limited to false claims. It is the absence of anyone who can answer for the words. A person who hands survey questions to a chatbot still receives the incentive, yet the responses may not reflect that person’s judgment. Teachers face essays whose authors cannot explain or revise the thinking the assignment was meant to develop. In each case, the record looks complete, yet the chain that once ran from words to a responsible speaker has broken.
Standing, in the sense used here, is the right for speech to count in a consequential process such as rulemaking, polling, or classroom evaluation. That right depends on answerability: the capacity of a named person or institution to explain, correct, or bear the consequences of the words. A witness under oath has both. An unnamed source borrows standing through a reporter who can be questioned. A chatbot has neither. It can produce fluent apologies and revisions, but nothing of its own is at stake, and operators can alter or replace the system at any time.
Assistance from these tools can remain legitimate when a person sets the purpose, reviews the output, adopts it, and remains able to defend the result. The failure occurs when the machine substitutes for the engagement the process requires. A student who submits generated work has not performed the intellectual work the task exists to develop. A flood of machine-generated comments that claim to represent separate individuals does not become legitimate simply because one operator can defend the set.
Practical redesigns focus on process rather than on making machines more humanlike. Named humans must review and adopt AI-assisted summaries or clusters, with intermediate steps open to inspection so corrections can propagate. Provenance records should identify which system performed each step and who approved the result, scaled to the sensitivity of the speech act. Institutions can verify a participant without public exposure when the role allows it, while requiring full identification for high-volume or powerful actors. The goal is to widen the number of people who can hold accountable roles, not to concentrate judgment among those who operate the tools.
Without such anchors, the volume of speech rises while the share that anyone can stand behind declines. Civic processes that accept unattributable input risk treating machine-generated language as equivalent to human judgment. Preserving the conditions under which words bind requires that responsibility remain with people who can answer for them.
Original analysis inspired by Deb Roy from Bulletin of the Atomic Scientists. Additional research and verification conducted through multiple sources.
