Neuron's AI Policy: Cell Press Rules for Neuroscience Authors

fMRI, PET, EEG, and MEG data can contain identifying information, particularly structural MRI scans, which can be used to reconstruct facial features. Many neuroimaging datasets are governed by strict data use agreements.

Don't input patient or participant neuroimaging data into cloud-based AI tools. This applies even to preprocessed or "anonymized" data, because:

• Cloud AI tools may store inputs for model training
• Structural MRI data can be re-identified even after defacing
• fMRI activation maps combined with demographic data may identify individuals
• Institutional data governance policies typically restrict data processing to approved systems

If your Neuron paper involves human neuroimaging, the AI disclosure should explicitly confirm that no participant data was processed through external AI tools.

Neuron papers frequently use AI-based research tools:

Research tools (standard STAR Methods): DeepLabCut (pose estimation), Suite2p (calcium imaging analysis), CaImAn (calcium imaging), FreeSurfer (brain segmentation), fMRIPrep (preprocessing pipeline), ANTs (image registration)

AI writing/code tools (STAR Methods AI disclosure): ChatGPT for language editing, Copilot for writing analysis scripts, Claude for restructuring text

The first category belongs in your standard computational methods description. The second requires a separate AI manuscript preparation disclosure. Don't conflate them, Neuron's reviewers include computational neuroscientists who will notice if your disclosure is unclear about which AI did what.

If your paper describes an AI model for neural decoding, brain-computer interface control, or neural signal processing, that's your research subject, it isn't covered by the manuscript preparation policy. Describe it fully in STAR Methods as methodology. Your ChatGPT or Copilot usage for writing goes in a separate disclosure.

This is especially important at Neuron because the journal publishes both the computational methods and their neuroscientific applications. A paper on a new decoder architecture needs two distinct sections: the decoder's technical description (research) and the writing tool disclosure (if applicable).

For a systems neuroscience paper:

"Two-photon calcium imaging data was analyzed using Suite2p (Pachitariu et al., 2017) as described in STAR Methods: Calcium Imaging Analysis. During manuscript preparation, the authors used ChatGPT (GPT-4, OpenAI) to improve the clarity of the Discussion section. GitHub Copilot (Microsoft) assisted with writing Python scripts for the population decoding analysis. All code was validated against manually computed results on a subset of recording sessions. The authors take full responsibility for the published content."

For a human neuroimaging paper:

"fMRI data was preprocessed using fMRIPrep v23.1 and analyzed using FSL and SPM12 (see STAR Methods: fMRI Analysis). No participant data was processed through any cloud-based AI tool. During manuscript preparation, Claude (Claude 3.5, Anthropic) was used to improve the language of the Introduction. All AI-suggested edits were reviewed by the corresponding author (M.L.). The authors take full responsibility for the content."

For a molecular/cellular neuroscience paper:

"During preparation of this manuscript, the authors used ChatGPT (GPT-4, OpenAI) to edit the Results section for conciseness. All content was verified against the experimental data by the senior author (K.P.). The authors take full responsibility for the published content."

The brain region illustration point matters. Neuron papers often include anatomical diagrams showing brain regions, circuit connectivity, or experimental targeting. These must be created with standard illustration tools (BioRender, Illustrator, Allen Brain Atlas templates) or derived from actual imaging data. They can't be generated by DALL-E or Midjourney, even if you plan to redraw them afterward. The generative origin disqualifies the image, it doesn't matter how much you edit it afterward.

Cell Press follows standard COPE-guided enforcement:

During review: Request to add disclosure. Neuron's reviewers include computational experts who may flag AI-generated patterns in code or text.

After publication:

• Correction for minor language editing non-disclosure
• Expression of concern if AI affected data analysis or interpretation
• Retraction for fabricated data or false claims

The clinical neuroscience concern: Some Neuron papers have translational implications for neurological and psychiatric disorders. If AI tools influenced how clinical data was interpreted or how treatment-relevant findings were described, the consequences extend beyond publication ethics into clinical responsibility.

Community dynamics: Neuron publishes roughly 200-250 articles per year. The neuroscience community at this level is tightly connected, many authors serve as reviewers for each other. A publication ethics issue at Neuron circulates quickly through conference networks and departmental channels.

Cell Press formalized its AI policy in early 2023:

The policy has been consistent since launch. Unlike AAAS (which initially banned AI text before reversing to a disclosure model), Cell Press went straight to disclosure from the start. For neuroscience labs with multi-year projects in the pipeline, this stability means the rules you're planning for today will almost certainly be the rules in effect when you submit.

Nature Neuroscience uses free-form Methods (Springer Nature style); Neuron uses STAR Methods (Cell Press style). The substantive AI requirements are identical. If you're preparing a manuscript for both as backup options, the disclosure content is the same, only the formatting differs.

Brain (Oxford UP) has especially high sensitivity to patient data given its focus on clinical neurology. The Journal of Neuroscience (SfN) is the society journal with the broadest readership in the field; its AI policy follows Oxford UP guidelines.

For electrophysiology and imaging papers:

• If AI helped write analysis code for spike sorting, calcium trace extraction, or connectivity analysis, disclose and validate
• Keep research tools (Kilosort, Suite2p, DeepLabCut) in standard STAR Methods
• AI-assisted code goes in a separate paragraph in Method Details

For human neuroimaging and clinical neuroscience:

• Never process participant data through cloud AI tools
• Explicitly state in your disclosure that patient data wasn't AI-processed
• Keep AI away from clinical interpretation sections

For computational neuroscience:

• If your paper develops an AI model, clearly separate the model description from any writing AI disclosure
• Deposit code in a public repository, Neuron expects code availability for computational papers
• AI-generated code should be independently validated

For behavioral neuroscience:

• If AI helped with behavioral analysis code (video tracking, classification), disclose and describe validation
• Behavioral scoring criteria should be human-defined, even if the scoring is automated

Before submission checklist:

• [ ] AI disclosure in STAR Methods → Method Details
• [ ] Research tools in standard STAR Methods
• [ ] Writing/code tools in separate AI disclosure
• [ ] No participant data processed through cloud AI
• [ ] No generative AI images or brain diagrams
• [ ] All analysis code validated and deposited
• [ ] All co-authors reviewed the disclosure

A Neuron submission readiness check can help verify your Neuron submission meets Cell Press standards before submission.

Comply proactively if:

• You used any AI tool (ChatGPT, Grammarly, Copilot) during manuscript preparation
• The journal requires AI use disclosure in the methods or acknowledgments
• Your institution has its own AI use policy that may be stricter

Less concerned if:

• You used AI only for grammar/spell checking (most journals exempt this)
• The journal does not have a formal AI policy yet
• Your use was limited to literature search or reference management