AI disclosure requirements for academic publishing depend on the publisher’s rules and how an author uses AI. The reported resignation of Dartmouth College Provost Santiago Schnell shows why that distinction matters. Authors should explain AI’s role, check its output, and follow disclosure policies rather than assume that editing assistance needs no acknowledgment.

Key takeaways

• According to the source report, Santiago Schnell resigned after a committee found failures to disclose significant AI use.
• The committee reviewed Dartmouth Claude logs but lacked access to relevant personal account logs.
• Schnell described AI as an editorial aid, while the committee said his use went beyond his public descriptions.
• Dartmouth continued promoting AI while calling for transparency and human judgment from its leaders.

The dispute puts a practical question before researchers: when does help from a chatbot become something readers must know about? The supplied Ars Technica report describes a case centered on disclosure, incomplete evidence, and the gap between an author’s account and records available to investigators.

What do AI disclosure requirements for academic publishing mean?

AI disclosure requirements for academic publishing mean authors must explain their AI use when the relevant publisher requires it. Calling a tool an editor, reviewer, or discussion partner does not by itself settle whether disclosure is necessary.

In Schnell’s case, the source reports that publishers required disclosure for the academic paper and Washington Post op-ed under review. The advisory committee found that he failed to disclose significant AI use in both.

That finding does not establish one rule for every journal or every writing tool. It does show why authors should check the policy for each publication instead of relying on their own definition of “editing.”

Responsibility and disclosure answer different questions

Responsibility concerns who stands behind the work. Disclosure explains how the work came together. An author can accept full responsibility for a manuscript and still fail to provide information a publisher requires.

Schnell’s own writing supports the first principle: he argues that human authors should take responsibility for the final text. The controversy concerns the second principle—whether he gave publishers the required account of AI’s involvement.

What did the committee find about Schnell’s AI use?

The committee found failures to disclose AI use, while acknowledging that it could not determine the full scope of that use, according to the source report.

On August 9, 2026, Schnell published the Washington Post op-ed “Universities are fighting AI cheating. But there’s a deeper problem.” His subsequent resignation brought scrutiny to his own writing practices.

The committee described his participation in its investigation as “insufficient cooperation.” Schnell supplied his Dartmouth laptop, but he did not provide relevant logs from his personal LLM accounts or agree to meet with the committee.

Investigators did obtain access to his Dartmouth Claude enterprise account. Those records showed him using AI to revise papers, identify and fix code bugs, and edit or generate figures.

The available records leave limits on the findings

The missing account logs matter because they left investigators without a complete record. They do not, on their own, establish what happened in those accounts.

The committee nevertheless said the records it reviewed went beyond Schnell’s public descriptions. He had told Semafor that chatbots helped him “refine a few arguments” and “check for grammatical errors.” The committee believed those descriptions might understate the work AI performed.

Where does editing end and substantive assistance begin?

The distinction becomes less clear when AI moves from suggesting wording to changing code, figures, or arguments.

One recorded request asked Claude to act as a reviewer with expertise in nonlinear chemical dynamics and mathematical chemical kinetics. Schnell wanted a referee report before submitting a paper, supplementary materials, and a repository to ChemSystemsChem.

That request differs from asking a chatbot to write an entire paper. It resembles seeking critical feedback. Other requests, however, asked Claude to make corrections directly to TeX files and the repository, and to generate new figures if needed.

For authors applying AI disclosure requirements for academic publishing, the useful question is not simply whether AI “wrote” the paper. It is what tasks the system performed and what the publisher asks authors to report.

Another exchange raised questions about intent. Schnell asked Claude whether a manuscript titled “Academic Freedom and Catholic Mission: The Procedural Overlap Zone in University Governance” contained “many signatures of AI writing.” The committee said he later submitted revisions incorporating the kinds of changes Claude suggested. That exchange raises concerns, but does not alone prove an effort to conceal AI use.

What should a useful AI disclosure explain?

A useful disclosure should describe the tool’s contribution and the author’s checks in terms a reader can understand.

Schnell offered a detailed example in his September 2026 paper “Who Wrote This?” According to the supplied source, he described using OpenAI’s GPT-5.6 Pro to explore the question, propose structures and titles, identify literature, and generate parts of early drafts.

He also described his own work: checking sources, correcting errors, rewriting sections, choosing examples, and making final decisions. He reported additional language assistance from GPT-5.6 Sol and critical reviews from GPT-5.6 Pro and Anthropic Claude Fable 5.1.

That account suggests a practical checklist, although it does not replace any publisher’s policy:

• Identify the AI tools used and the tasks they performed.
• Distinguish feedback and language editing from generated drafts, code changes, or figure work.
• Explain which sources, claims, calculations, and outputs the author checked.
• Describe the author’s revisions and responsibility for the final manuscript.

Schnell rejected “coauthored” as the right label, arguing that contribution and authorship are different roles. Whatever label authors choose, a clear account of the work gives publishers more useful information than a broad statement that AI merely “helped.”

Why does Dartmouth’s response matter beyond one author?

Dartmouth’s response separates support for AI from permission to use it without transparency.

The university continues to promote AI research and use. Its “What Must Remain Human?” conference includes an Anthropic-sponsored student hackathon and speakers such as Mira Murati and Yann LeCun.

President Sian Leah Beilock also wrote “Why I Want More AI at Dartmouth” for The Atlantic in September. In announcing Schnell’s resignation, she stressed that leaders must model the transparency and human judgment expected from students and scholars.

The student survey described in the source shows why clear expectations matter. While 61 percent said AI helps explain difficult concepts, 70 percent considered take-home work unfair because of AI. Another 39 percent felt pressure to use it because peers did. These findings concern classroom work rather than journal policies, but they reveal the pressure created by uncertain boundaries.

Our take

This case suggests that promoting AI and enforcing disclosure are compatible, but institutions must explain both clearly. The new development is not simply another argument about AI cheating: it is a reported leadership resignation over disclosure failures while Dartmouth continues expanding its AI activities.

Researchers and publishers face the most direct consequences. For AI companies and developers, the case highlights questions about documenting tool use. For consumers of research, including readers in Germany and the EU, the issue is whether published work offers a clear account of how it was produced. The source establishes no separate European requirement.

One open question is how investigators should seek relevant personal-account records while protecting unrelated private material. The source does not resolve that balance. In the coming weeks, readers should watch for clearer publisher guidance and any further evidence that narrows the uncertainty about Schnell’s use. Disclosure is most useful when it describes actual tasks, not just tool names.

For researchers, AI disclosure requirements for academic publishing should be part of the writing process, not a last-minute submission check. Read the publisher’s policy, record meaningful AI assistance, and explain both the tool’s contribution and your own review.

(Source: https://arstechnica.com/tech-policy/2026/10/dartmouths-provost-warned-us-about-ai-cheating-but-his-own-ai-use-cost-him-his-job/)

For more news: Click Here

FAQ