A group of climate scientists who wrote a critical review of climate science used large language models to edit and assess their work, according to the Bulletin of the Atomic Scientists. The exchanges show AI being used not just to polish prose, but to challenge the group’s claims and help prepare for criticism.
Watch Desk analysis
What happened
The Bulletin says emails released publicly by the Environmental Defense Fund and the Union of Concerned Scientists revealed how five members of the US Climate Working Group used AI tools while preparing a report. The documents included requests to Microsoft Copilot to assess scientific accuracy and potential bias, shorten an executive summary, and draft a rebuttal opinion piece.
Copilot reportedly flagged cherry-picked evidence, narrow sourcing and language it considered rhetorically loaded. The group discussed editing some wording in response. Members also used AI for other tasks, including rephrasing material and composing poems. One member, Ross McKitrick, told the Bulletin that the group did its own research and writing and did not delegate the report to AI.
Why it matters
This is a revealing example of AI acting as an informal reviewer inside a politically contested scientific process. The tools were not simply asked to tidy sentences: their comments fed into discussions about how the report should present its arguments and anticipate objections.
That does not make the models independent peer reviewers, or settle whether their criticisms were right. But it does show how AI feedback can shape the way authors frame work, even when people retain responsibility for the final text.
Our read
The interesting detail is not that a model spotted words with a tone problem. It is that the group treated its feedback as worth discussing, while also questioning whether the models favoured consensus sources and misunderstood dissenting arguments. AI can be a useful critic, but it is a critic with its own limits, not an impartial referee in a box.
The Bulletin reports that it found no evidence the group used AI to generate the report’s substantive content or citations wholesale. That distinction matters: assistance with review and editing is not the same claim as machine-written science.
What to watch
- Whether the released emails prompt further scrutiny of the report’s sources and methods.
- How research groups disclose AI use in reviewing, editing and drafting.
- Whether AI feedback is checked against evidence, rather than treated as a verdict.
Discussion spark: Should researchers disclose when AI feedback influences how they frame a report, even if people wrote and researched the final work themselves?
Sources and evidence
- Climate skeptics used AI to review their work. It told them they were biased. – Bulletin of the Atomic Scientists (31 October 2026, 10:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.