Researchers measure the carbon cost of AI-written research papers for the first time

A software architecture conference just published the first carbon audit of AI-generated text in research papers, plus the environmental cost of the conference itself. This means institutions now have actual numbers—not guesses—about how much computational energy goes into AI writing and conference operations, which creates pressure to measure and reduce both.

For years, research institutions treated AI tool usage as invisible—no tracking, no carbon accounting, no visibility into what the computational cost actually was. This paper puts numbers on it, which means conference organizers and researchers can no longer claim they don't know what they're burning. The second-order effect: once you measure something, you're implicitly responsible for it. Expect software architecture conferences to start asking authors to disclose AI usage the way they disclose conflicts of interest.