Libby's AI Filters Depend on Publishers Self-Labeling
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Libby is rolling out AI content filters covering AI authorship, AI-narrated audiobooks, machine translation, and AI-generated art, letting readers and librarians opt out via app settings.
- Marc DeBevoise, OverDrive's new CEO who took over last week, said the filters rely on publishers self-labeling through standardized metadata — OverDrive decided against using any AI detection tool.
- Libby's catalog spans over 6 million books borrowed more than 1 billion times across 92,000 libraries, schools, and universities in 115+ countries; DeBevoise noted "everything before 2020 [or] 2022 is by definition not AI."
- Draft2Digital, the self-publishing intermediary that supplies books to Apple Books and Google Play Books, allows AI-generated titles as long as they've undergone "extensive editing from a human" — making it "all but inevitable" some AI works reach Libby.
- Kobo CEO Michael Tamblyn revealed last month that Kobo rejects nearly half of all self-published books over AI concerns, calling the pipeline "a firehose"; Amazon restricted daily self-pub uploads in 2023 for the same reason.
- Audiobooks make up only 15% of Libby's catalog but roughly 50% of its usage, and DeBevoise sees AI localization (domestic to international and back) as a major opportunity, while still preferring human voice actors for new recordings.
- Libby's machine-translation filter only functions if books are correctly labeled — a gap critics have flagged as especially problematic for literary works, where AI translation often misses nuance.
Why it matters: The self-labeling approach has an obvious weak link: Draft2Digital already allows AI-assisted books after human editing, and OverDrive concedes some AI titles will inevitably reach Libby's catalog. With 6 million titles and 1 billion borrows, even a trickle of mislabeled AI books means readers who opted out will still get them — undermining the filter's core promise.
