Springer Nature Now Judges AI Use by How It Is Used, Not Whether

Most publisher AI policies are built as a list: this is allowed, that is not, declare the rest. Springer Nature’s is built as a risk assessment, sorting uses by how much of the intellectual work the model is doing and how visible that is to a reader.

The author-facing rules that follow from it are short. Disclose the use, “including model and purpose of use”. Keep the argument, the interpretation and the conclusions author-led. And stay answerable: “you remain fully accountable for everything submitted in your name”.

Images get a second disclosure. Visual content “that has been created, substantially modified or enhanced using AI should be disclosed clearly and transparently”, and “specific disclosure should be provided within the caption or legend of the visual content”. A global AI statement at the end of the paper does not discharge it — the disclosure has to sit where the reader meets the image.

Reviewers. The peer review page is firmer still. “Manuscript text, figures, or identifiable content should not be uploaded into these types of tools.” The review “must remain clearly human-led, defensible, and accountable”, reviewers “must independently verify all outputs”, and AI “must not replace expert evaluation”.

A binary rule ages badly. “Declare if you used AI” was workable when AI meant a chatbot that drafted paragraphs; it is close to meaningless when the same label covers a reference manager, a translation tool, a statistical package and a model that wrote the discussion. Sorting by how much of the thinking was delegated keeps the policy pointed at the thing that actually matters.

The caption-level image disclosure follows the same reasoning about where harm occurs. A reader forms a belief from a figure in the second they look at it, and a disclosure three pages later arrives after that belief has formed.

One thing to be straight about: neither page carries a date. For a policy whose whole subject is moving quickly, that is a real weakness, and it is the reason this brief says “no date on the page” rather than borrowing the footer year. Check it yourself before you rely on it.

How it reaches a specific journal. Springer Nature publishes the framework at group level and mirrors it onto journal editorial-policy pages, including Nature Portfolio titles. The wording on a journal page may be shorter; the obligations are the same ones.

A practical consequence for authors submitting to Nature-branded journals: the caption-level image disclosure is easy to miss, because most other publishers ask only for a single statement. Check your figure captions before submission rather than at proof.