Science Advances' AI Policy: AAAS Rules for the Multidisciplinary OA Journal

Understanding how the AAAS got here explains the current enforcement culture:

The initial ban was significant because it was the strongest position any major publisher took. Most competitors (Springer Nature, Cell Press, Elsevier) went straight to disclosure models. AAAS had to walk back a more restrictive policy, which created institutional memory within the editorial team about AI's impact on manuscript integrity.

The practical implication: AAAS editors are arguably more attuned to AI in manuscripts than editors at publishers that never banned it. They had ten months of actively looking for AI-generated text during the ban period, and that vigilance didn't disappear when the policy changed.

AAAS directs AI disclosure to the Acknowledgments section and/or Methods. For Science Advances, where Methods sections are often detailed, the best practice is to include it in both:

Acknowledgments disclosure:

"The authors acknowledge the use of ChatGPT (GPT-4, OpenAI) for language editing during manuscript preparation."

Methods section disclosure (more detailed):

"During preparation of this manuscript, the authors used ChatGPT (GPT-4, OpenAI) to improve the clarity and readability of the Discussion section. GitHub Copilot (Microsoft) was used to assist with writing Python scripts for the molecular dynamics analysis. All AI-generated text suggestions were reviewed by the corresponding author (R.T.), and all code was validated against published benchmarks. The authors take full responsibility for the published content."

What wouldn't pass:

"AI tools assisted with this work."

Too vague. Doesn't identify the tool, the purpose, or the scope. AAAS editors will request specifics.

For multidisciplinary submissions

Science Advances covers everything from astrophysics to zoology. The disclosure format should reflect your field's conventions:

For a materials science paper:

"The authors used Claude (Claude 3.5, Anthropic) to edit the Introduction for language clarity. Density functional theory calculations were performed using VASP 6.3 without AI assistance. The authors take full responsibility for the content."

For a biological sciences paper:

"ChatGPT (GPT-4, OpenAI) was used to improve the readability of the Results and Discussion sections. No AI tools were used for experimental design, data collection, or biological interpretation. All AI-suggested edits were reviewed by three co-authors (J.L., M.K., S.P.)."

For a computational science paper:

"The neural network described in this paper was developed using PyTorch 2.1 (see Methods: Model Architecture). Separately, during manuscript preparation, the authors used ChatGPT (GPT-4, OpenAI) to edit the Introduction. The research methodology and the editing tool are entirely separate. The authors take full responsibility for the content."

AAAS enforcement reflects the organization's history with AI policy:

During review:

• Editor contacts corresponding author
• Disclosure must be added
• Given the AAAS's enforcement history (ten months of active ban), editors are particularly vigilant
• Deliberate concealment can lead to rejection with prejudice

After publication:

• Correction for minor undisclosed language editing
• Expression of concern for unclear scope
• Retraction for fabricated content or false claims
• COPE investigation for systematic issues

The AAAS reputation factor: Science and Science Advances are published by the world's largest general scientific society. A publication ethics issue at either journal attracts attention from the broader scientific community, science journalists, and policy makers. The AAAS's annual meeting draws 10,000+ attendees, and publication ethics discussions at AAAS events reach a wider audience than at most scientific conferences.

Science Advances is the only journal in this comparison whose publisher previously banned AI-generated text. This history makes its current disclosure requirements somewhat more enforced in practice than journals that started with disclosure models.

PNAS requires disclosure in both Methods and Author Contributions, making it the most prescriptive about placement. eLife's funder-backed model means it has lower financial pressure around volume than fully APC-dependent publishers, which may affect how aggressively it enforces AI compliance.

For all submissions:

• Disclose in both Acknowledgments and Methods. AAAS guidelines mention Acknowledgments, but a detailed Methods disclosure provides the specifics reviewers need.
• Remember that AAAS editors have institutional memory of the ban period. They're trained to notice AI-generated language patterns.
• If you're a non-native English speaker who used AI for language editing, this is perfectly acceptable, just disclose it transparently.

For papers under 4,000 words:

• Keep the disclosure concise but complete: tool name, version, purpose, review confirmation
• Don't sacrifice research content space for an overly long AI disclosure

For papers with extensive supplementary material:

• AI use in supplementary material text needs disclosure too
• If supplementary methods sections were AI-assisted, include this in the main Methods disclosure

For computational papers:

• Clearly separate research AI from writing AI
• Deposit code in a public repository
• AAAS expects data and code availability for computational work

Before submission checklist:

• [ ] AI disclosure in Acknowledgments section
• [ ] Detailed AI disclosure in Methods section
• [ ] Tool name, version, and use case specified
• [ ] No AI-generated images
• [ ] Code deposited if applicable
• [ ] All co-authors aware of AI disclosure
• [ ] Research AI and writing AI clearly distinguished (if applicable)

A free manuscript assessment can help verify your Science Advances submission meets AAAS standards before submission.