What is new is an essay, not a detector
MIT News published a Q&A with Deb Roy today. He is a professor of media arts and sciences at MIT and director of the Center for Constructive Communication. The new essay is in the Bulletin of the Atomic Scientists, titled "Who stands behind the words?" His argument is that public communication still needs a person who is accountable and able to answer, even when AI drafted the text.
Answerability, in his usage, is not a new statute. He borrows the idea that a witness has to face questions: a person with standing to speak also has to answer for what they say. He writes that this does not require a name on every sentence. An institution can verify a person and stand behind the words without exposing them. An accusation asks more. Someone trusted, such as a reporter, has to be able to know who is speaking.
[1]
The opening number is a 2017 case
Roy opens with the Federal Communications Commission's 2017 net-neutrality comments. The commission received more than 22 million public comments, and the New York attorney general later found that nearly 18 million of them were fake. He writes that the fraud was caught because it left fingerprints: templated text and duplicated phrasing. Once large language models erase those fingerprints, every comment can be fluent and different, and reviewers can no longer tell which ones express views that real people formed and will stand behind.
That is his use of an old case, not a new 2026 count. The Q&A also does not present a detection system that has been deployed.
[1]His own center still lets people override the machine
The center, and Cortico, the nonprofit Roy co-founded, use tools including AI to find themes in discussions. The process is designed so participants can check and override the machine output at each key step. A person is responsible for every public comment that goes out.
His view of government reviewers is that they have quietly lowered the weight they give public comments. That is his sense. The Q&A does not cite a new government statistic showing that the comments are no longer read.
[1]