WASHINGTON — Researchers on a Trump climate panel asked Microsoft Copilot to review their draft report. It flagged selective evidence and rhetoric that weakened the draft’s scientific objectivity. The exchange appears in court-released emails circulated before the Energy Department published the assessment.

An Oct. 31 investigation by the Bulletin of the Atomic Scientists brought renewed attention to the group’s use of artificial intelligence. Economist and co-author Ross McKitrick told the publication that the researchers experimented with the technology but conducted their own research and writing. He rejected the suggestion that they delegated authorship to an AI system.

The records illuminate more than an awkward encounter with a chatbot. They concern an official government report, a review process that a federal judge found violated transparency requirements, and scientific conclusions that researchers have challenged independently. The AI feedback is one part of that record, not the final authority on whether the report’s claims hold up.

What the Trump Climate Panel Published

The Energy Department released the report July 29, 2025, under the title “A Critical Review of Impacts of Greenhouse Gas Emissions on the U.S. Climate.” Energy Secretary Chris Wright assembled the five-member Climate Working Group to assess research on greenhouse gases and their consequences. The department described the authors as independent scientists with expertise spanning climate research, physical science and economics.

The report argued that warming caused by carbon dioxide appeared less economically damaging than commonly believed. It also suggested that aggressive emissions-reduction policies could cause more harm than benefit. Those conclusions challenged the rationale for substantial government action to limit greenhouse gas emissions.

This was therefore more consequential than a private disagreement among researchers. The department published and promoted the assessment through official channels while inviting public comment. Its presentation gave the authors’ conclusions the visibility of a federal scientific publication, making both their evidence and their review process matters of public interest.

What AI Flagged in the Trump Climate Panel Draft

On July 10, 2025, physicist Steven Koonin circulated Copilot’s assessment to the group. The tool said the draft emphasized carbon dioxide’s benefits while underplaying heat stress, nutrient limitations and other constraints. It also credited some discussions of scientific uncertainty rather than dismissing the entire draft.

“This is amazing, far better than what we would get from ‘real’ scientists,” climate scientist Judith Curry wrote. She said she was examining the draft for “Skeptic talk” and red-flag wording. Her response welcomed the critique while identifying language as an immediate editing concern.

Those comments document attention to presentation, but they do not establish which substantive revisions followed. Determining whether the authors adequately addressed a particular criticism requires comparing the draft, underlying evidence and final text. Removing loaded wording and correcting an imbalance in the evidence are different tasks.

The Chatbot Also Needed Fact-Checking

The AI response itself illustrates why its conclusions cannot substitute for scientific review. In one passage, Copilot accepted the argument that calling the process ocean acidification was misleading because seawater remains alkaline. That reasoning confuses a change in acidity with whether water has crossed the threshold into an acidic condition.

As the National Oceanic and Atmospheric Administration explains, ocean acidification refers to a sustained reduction in ocean pH. Absorbed carbon dioxide changes seawater chemistry, increasing hydrogen ions and reducing carbonate availability. Water does not need to fall below neutral pH for that process to occur.

That error does not invalidate every criticism Copilot offered. It does demonstrate why the exchange cannot responsibly be presented as a machine conclusively settling climate science. The defensible question is whether each claim survives scrutiny against measurements, published research and the relevant scientific methods.

Scientists Challenged the Report With Data

Separate from the AI exchange, researchers have identified specific problems in the report’s treatment of atmospheric temperatures. An analysis published in Earth’s Future on Aug. 29, led by climate scientist Benjamin Santer, examined its claims about the human influence on climate. The authors challenged the report’s suggestion that recent atmospheric temperature patterns were inconsistent with that influence.

The dispute concerns a recognizable pattern: warming in the lower atmosphere alongside cooling higher in the stratosphere. Santer and his co-authors found that an apparent basis for the report’s argument relied on incorrect observational trends. Using the appropriate observations, they identified the expected human influence with high confidence.

The researchers also criticized the absence of supporting references for the report’s disputed statements. They called for scientific assessments to provide traceable data, clear sourcing and independent oversight of responses to reviewers. Unlike the chatbot’s broad assessment of tone and balance, their analysis tested a specific scientific claim.

Trump Climate Panel Violated Transparency Requirements

The report’s preparation also faced a separate legal challenge. In a Jan. 30, 2026, judgment, U.S. District Judge William G. Young determined that the Climate Working Group was subject to the Federal Advisory Committee Act. The government had argued that the group merely exchanged information with federal officials and therefore qualified for an exemption.

Young rejected that characterization because the group provided substantive policy advice and recommendations. The judgment established violations involving committee procedures, open meetings, public records, and requirements for fair balance and influence. It also noted that the government had remedied the plaintiffs’ informational injury by producing records.

The ruling concerned how the government obtained and handled advice, not a judicial determination that every scientific claim was false. Young dismissed the Environmental Protection Agency as a defendant after finding no persuasive evidence of violations outside the Energy Department and Wright. Those distinctions separate the court’s findings from broader criticism of the administration’s climate policies.

These Are Previously Released Records

The Environmental Defense Fund and Union of Concerned Scientists obtained the documents through their lawsuit. Their public records collection identifies an initial release in January 2026 and the broader publication of records in March. The October reporting examines material from that disclosure rather than a report written or emails released this week.

Both organizations opposed the working group’s approach and describe its report in sharply critical terms. Their advocacy should be distinguished from the contents of the documents they made available. The emails and judgment allow readers to examine specific exchanges and legal findings without relying solely on the organizations’ characterizations.

The chronology also matters when assessing the government’s response. The report dates to July 2025, the judgment to January 2026 and the latest Bulletin investigation to October. Treating those developments as one newly unfolding event would obscure when officials received criticism and when the public gained access to the underlying record.

EPA Disclaimed Reliance on the Report

The report emerged during the administration’s effort to overturn the EPA’s 2009 greenhouse gas endangerment finding. That finding established that specified greenhouse gases threatened public health and welfare under the relevant Clean Air Act framework. USA Herald previously covered questions about that regulatory agenda during Lee Zeldin’s confirmation process.

However, the EPA’s final rule published Feb. 18, 2026, expressly disclaimed reliance on the working group’s assessment. The agency said neither the May draft nor the July report supported any aspect of its final action. It instead advanced legal arguments about its authority under the Clean Air Act and other grounds described in the rule.

That distinction limits what can be claimed about the report’s regulatory role. It formed part of the earlier policy debate, but describing it as the scientific basis for the final repeal would contradict the rule’s stated rationale. Criticism of the report’s preparation does not require overstating how the EPA ultimately used it.

The Scientific Stakes Extend Beyond the Emails

The broader evidence also deserves attention independent of the working group. A 2025 assessment by the National Academies of Sciences, Engineering, and Medicine concluded that the scientific basis for the 2009 endangerment finding had strengthened. Longer observational records and additional research reinforced evidence of harm from human-caused greenhouse gas emissions.

The assessment described risks involving extreme heat, wildfire smoke, other air pollution and severe weather. It also documented consequences for agriculture, infrastructure, water systems and ecosystems. Its authors distinguished areas of continuing uncertainty from conclusions supported by multiple lines of evidence.

Those findings explain why the dispute is not merely about academic tone. Assessments of climate risks inform decisions affecting public health, infrastructure and economic costs. Evaluating them requires attention to the full body of evidence, including limitations and contrary findings, rather than selecting whichever authority produces the preferred answer.

The Unresolved Question Is How Criticism Was Addressed

McKitrick’s response to the Bulletin places an important limit on the story: The authors maintain that they performed the research and writing themselves. Experimenting with AI does not, on its own, establish scientific misconduct. Nor does asking a tool to identify weaknesses demonstrate that the researchers intended to ignore them.

The more substantive issue is how the report’s conclusions withstand examination after those weaknesses are identified. The published scientific critique points to claims that can be tested against observational records. The court judgment establishes failures in the advisory process. Neither depends on treating a chatbot’s response as an expert verdict.

Together, those records shift the focus from whether AI embarrassed a group of researchers to whether an official assessment met defensible standards. A more neutral tone cannot substitute for complete evidence, just as an AI-generated criticism cannot substitute for verification. The report’s credibility ultimately rests on its data, methods and documented answers to substantive objections.