IEEE Generative AI Policy

IEEE does not publish one single, centrally-branded ‘generative AI policy’ document the way some publishers do. Instead, consistent generative-AI rules are disseminated across IEEE’s journals, transactions, and conferences through IEEE Author Center guidance and through parallel guidance pages maintained by individual IEEE technical societies — among them the IEEE Aerospace and Electronic Systems Society (AESS), Circuits and Systems Society (CAS), Robotics and Automation Society (RAS), and Power & Energy Society (PES). The substance is consistent society to society, but authors submitting to a specific IEEE venue should still confirm the exact wording in that venue’s own call for papers or author guidelines, the same caveat that applies to IEEE’s peer-review model, which is likewise set at the society level rather than uniformly across IEEE as a whole.

The policy splits into two genuinely distinct concerns that are easy to conflate but govern different people at different stages of the editorial process: what authors must disclose when they use generative AI to help prepare a submission, and a separate, stricter set of rules governing what reviewers may and may not do with AI tools while evaluating a confidential manuscript.

Author-side rules: disclosure and accountability

For authors, IEEE’s guidance follows the pattern now common across major publishers, closely paralleling the general generative-AI disclosure statement convention:

• Disclosure is required, not optional. Where generative AI tools (large language models such as ChatGPT, Microsoft Copilot, Google Gemini, or similar assistive writing tools) are used to help prepare submission text, that use must be disclosed, typically in the acknowledgments section of the paper.
• No AI co-authorship. An AI system cannot be listed as an author or co-author on an IEEE paper. Authorship implies accountability that a tool cannot bear.
• Full human accountability. Authors remain 100% responsible for the accuracy, originality, and integrity of every word, figure, chart, and citation in the paper, regardless of whether AI assisted in drafting it. If AI-generated text introduces plagiarism, fabricated citations, or factual errors, that is treated as the authors’ error, not the tool’s.
• No fabrication via AI. Using generative AI to fabricate or manipulate data, figures, or code is prohibited outright — this is a research-integrity violation independent of the disclosure question, not a milder version of it.

Reviewer-side rules: a confidentiality problem, not a disclosure problem

The reviewer-facing rules are stricter and rest on a different rationale entirely. A manuscript under review is confidential, unpublished material entrusted to a reviewer under the expectation that it won’t circulate before publication. Uploading any portion of that manuscript to a public generative-AI platform risks exactly that: most public AI systems retain, and may train on, whatever text is submitted to them, meaning a reviewer who pastes a manuscript excerpt into a public chatbot has potentially disclosed unpublished research to a third-party system outside the authors’ and publisher’s control.

Because of that, IEEE reviewer guidance draws a firm line:

• Reviewers may not upload manuscript content to public AI tools — processing any part of a manuscript under review through a public generative-AI platform is treated as a breach of confidentiality, independent of what the reviewer intended to do with the output.
• Reviewers may not use public AI platforms to generate part or all of a review’s substantive content — the scientific judgment in a peer review is expected to be the reviewer’s own.
• Narrow editing assistance is treated differently. Using an AI tool purely for grammar or wording polish of a reviewer’s own already-formed comments — without submitting manuscript content itself — falls outside the core prohibition; disclosure of that narrow use is recommended but not strictly required.

This author/reviewer split is not unique to IEEE. Several other major publishers draw a comparable confidentiality-driven line for reviewers while handling author disclosure separately — see the AI in Peer Review: Publisher Policy Comparison page for how IEEE’s approach sits alongside Elsevier, Springer Nature, and other major publishers’ reviewer-AI rules side by side.

Worked examples

Example 1 — author disclosure, compliant. An author uses a commercial LLM to suggest phrasing improvements to an already-drafted results section, keeps all data, analysis, and conclusions their own, and adds a one-line acknowledgment noting the tool and its limited role. This satisfies the disclosure requirement; the AI is not listed as a co-author and the authors remain accountable for the substance.

Example 2 — reviewer confidentiality breach. A reviewer, wanting a faster first read of a long manuscript, pastes several paragraphs of the submission into a public AI chatbot and asks it to summarize the contribution and flag weaknesses. Even though the reviewer never intended to plagiarize or misuse the material, this act itself — exposing unpublished, confidential manuscript text to a third-party system — is the policy violation, regardless of what the reviewer does with the AI’s response.

Counter-example

A reviewer runs a completed, entirely self-written review through a grammar-checking AI tool to fix phrasing, without pasting in any manuscript text or manuscript-specific detail. Because no confidential manuscript content was submitted to the tool and the reviewer’s own judgment produced the substance of the review, this falls outside IEEE’s core reviewer prohibition, even though disclosure of the assistive tool is still good practice.