Tamil Nadu Library Probe Puts AI-Written Books Under Scrutiny

A new public library in Tamil Nadu, India, has become an unexpected test case for a problem that's hard to see from the outside: machine-written text ending up on shelves meant to be trusted. The allegations are unproven, but the way they surfaced points to where the weak spots really are.

The Periyar Arivulagam library cost 3 billion rupees (about $34 million, or £26 million) and opened on October 16. Almost immediately, the acclaimed Tamil novelist Perumal Murugan said as many as 1,800 of the books it had purchased may be partially or wholly generated using AI.

What's actually being alleged

According to BBC reporting, the books in question came from a single publisher and were bought at unusually high prices. Many contain false information, unfamiliar or suspicious author names, and ISBNs that don't hold up. Murugan raised the alarm in an online magazine after visiting the library. He described thousands of titles in the Tamil section that were written by people he didn't recognise, contained inaccurate details, and were priced well above what you'd expect.

That's a specific set of signals, and it's worth separating them. Bad facts and odd bylines can come from a sloppy human publisher just as easily as from a language model. What makes the AI question live is the pattern: high volume, one supplier, and errors that look like the kind a generator produces when it fills in plausible-sounding detail without checking it. Even so, it's still unclear whether AI was used to create or alter the published books at all. Anyone writing about this case should keep that uncertainty front and centre.

Why procurement is where this breaks

Here's the thing: a library's catalogue is a chain of trust. Selectors and acquisitions staff typically rely on publisher reputation, review coverage, and basic metadata checks such as whether an ISBN resolves to the right title. When a large share of purchases comes from a little-known imprint at inflated prices, that concentration is itself a warning sign. Procurement processes usually aren't built to read prose, so a text can pass every administrative check and still be riddled with errors.

The political fallout has followed the paperwork. The state administration has blamed the previous administration for the purchases, an allegation the Dravida Munnetra Kazhagam denies. Tamil Nadu's School Education Minister, A Rajmohan, said that if the accusations hold up, the episode could be "India's first AI scam." That framing is doing a lot of work, and it'll matter whether the eventual findings focus on how the books were produced or on how they were selected and paid for. Those are different failures with different fixes.

The scrutiny isn't confined to one building. Two other Indian libraries have also come under review after inaccuracies were found in their books, and the state has ordered an internal audit of public library collections. More than 160 writers, researchers, artists and publishers have signed a public statement calling for an independent inquiry into the Periyar Arivulagam library. They also want procurement records released publicly and an audit of collections across the state.

A wider pattern in books, schools and creative work

The library case lands in a broader debate about AI's effect on knowledge and creative industries. A recent study by Fizz found that 96% of US college students use AI. Supporters see that as a productivity gain. Critics worry it could erode cognitive skills and critical thinking, and some schools have chosen to opt out entirely in an effort to protect pre-AI standards. Google has pushed back on that approach, saying it will disable Search in schools that opt out because "Search and AI are deeply integrated."

On the creative side, the disputes are mostly about training data and likeness. Bestselling author David Baldacci has filed a lawsuit against OpenAI, claiming the company trained its models on his books without permission. The illustrator Axel Scheffler, known for The Gruffalo, has said he is "seriously considering" legal action against AI companies after seeing his character reproduced in posters and advertisements. These cases are about what gets ingested and what gets output, which is a different question from the one the Indian library raises about what gets published and shelved.

And detection is getting harder. Researchers have suggested that AI systems may eventually learn to correct their own writing style until the signals that detection tools rely on disappear. If that happens, a librarian or editor can't count on a classifier to flag suspect manuscripts. The checks have to move upstream, into publisher verification, provenance records and the kind of plain fact-checking that catches invented citations.

What to watch next

For anyone building catalogue, metadata or acquisitions tooling, the practical lessons are concrete. ISBN validation, cross-referencing author identities and flagging clusters of purchases from a single new imprint are all checkable without guessing at how a text was written. Those controls don't depend on detecting AI, which is useful because detection is unreliable and getting less so.

The open questions are the ones that will decide how this story reads. Was AI involved at all, or were these errors the ordinary result of a bad publisher? If an independent inquiry is held, its procurement findings will likely matter as much as any technical analysis of the text. Either way, a new library full of unvetted books is a reminder that the cheapest way to fill a shelf is often the most expensive way to lose its readers' trust.