YOUR GO-TO INTELLIGENCE RESOURCE FOR THE SCHOLARLY PUBLISHING INDUSTRY

Springer Nature has donated its proprietary AI tool for identifying AI-generated nonsensical content to the STM Integrity Hub, a cross-industry initiative focused on maintaining research publishing standards. Previously used internally across Springer Nature’s journals and books, the tool will now be available to publishers industry-wide.

The tool, originally named Geppetto, was developed by Springer Nature’s research integrity team in collaboration with Slimmer AI’s Science division, which the publisher acquired in 2023. It operates by dividing manuscripts into sections and applying proprietary algorithms to evaluate textual coherence and consistency. Each section is scored based on the likelihood of AI-generated content. High-probability scores trigger review by human assessors.

The tool has detected hundreds of fraudulent submissions early in the editorial workflow, preventing publication and reducing editorial burden. It has also assisted in uncovering clusters of problematic submissions linked by shared themes or special issues, enabling broader investigation into potentially compromised content.

By contributing the tool to the STM Integrity Hub, Springer Nature supports broader adoption of content screening technologies by publishers of all sizes. The STM Integrity Hub facilitates collaborative development of tools that identify indicators of fraudulent or manipulated manuscripts, including those generated by paper mills.

As part of its ongoing integrity initiatives, Springer Nature continues to develop additional AI-based tools such as Snappshot, which flags problematic images, and a recently launched reference checker. These efforts are supported by continued investment in expert teams, in-house technologies, and collaborative partnerships with publishing sector bodies.

Integrating the AI text analysis tool into the STM Integrity Hub aims to expand detection capabilities across the publishing community, improve training datasets through wider usage, and enhance overall accuracy in identifying suspect content.