James O'Donnell argues in MIT Technology Review that chatbots should be able to end conversations

James O'Donnell argues in MIT Technology Review that chatbots should be able to end conversations

MIT Technology Review’s James O'Donnell argues on October 21, 2025, that chatbots should be able to stop talking to users. He says chatbots can produce endless streams of humanlike, authoritative and helpful text, which can facilitate delusional spirals, worsen mental-health crises and otherwise harm vulnerable people. The proposed safeguard would let a chatbot end a conversation, even though that could reduce the time people spend using the product. O'Donnell asks why companies are not using it, but this newsletter excerpt does not give their reasons or explain how the safeguard would work. His argument centers on a tension between continued product use and the risk of harm to vulnerable users.

Publication: ca88679e-a9fb-e6f8-fd52-4e155489e1cd

attribution — pass: This newsletter issue includes editorial commentary, so it counts as one piece, not a bare listing. The reading draws only on the segment signed by James O'Donnell and correctly attributes the argument to him at MIT Technology Review.

So, why aren’t companies using this obvious safeguard? Read the full story. —James O'Donnell

summary — pass: The title, summary and question match the segment's argument: chatbots should be able to stop conversations because endless text can harm vulnerable people, despite reducing product use. The reading is in plain language, discloses that it relies on an excerpt, and makes its gaps clear. The question follows from the source's own unanswered question.

But the one thing that almost no chatbot will ever do is stop talking to you.

date — uncertain: October 21, 2025 matches the UTC day of evidence.originalPublished for this Download issue. However, the reading says O'Donnell argues this on that date, while the source says the segment originally appeared earlier in The Algorithm, on a date not given. His argument's first publication may precede October 21. A smaller wording repair would date the newsletter issue rather than the argument. The reading uses none of the later Deep Dive items.

This story originally appeared in The Algorithm, our weekly newsletter on AI.

support — pass: Each claim is source-stated with exact supporting excerpts. The text calls the feature a safeguard, ties it to reduced usage time, and lists harms to vulnerable people. The reading adds no unsupported benefits or reasons, and it notes that the excerpt does not explain why companies do not use the safeguard or how it would work.

AI’s ability to generate endless streams of humanlike, authoritative, and helpful text can facilitate delusional spirals, worsen mental-health crises, and otherwise harm vulnerable people.

media — pass: No media was selected, and no verified media candidates were supplied. That is acceptable.