AI

More than a dozen groups have asked the Federal Trade Commission to look at the scanning and destruction of books for the training of AI models.

Commentators find it problematic that the watermark gets applied when Claude is used for assistive tasks rather than generative tasks.

But breathless coverage misses the fact that most of these books may be out of date, undesirable to readers, and not valuable per se.

Inexperienced authors may believe generative AI use improves their manuscript, but there’s a cost—extra fees charged by freelance editors.

Due to EU transparency laws around AI-generated content, future models will add embedded watermarks and digitally signed provenance metadata.

Across 15 journalism categories, five award winners disclosed AI use in their submissions, the most since the disclosure was added in 2024.

Spotify will roll out an AI Persona badge that signals an artist’s identity may be AI generated and does not represent a real person.

A debut author had his $2+ million book deal canceled because his agents came to believe his manuscript was substantially AI assisted.

It covers AI-generated or AI-manipulated content that could be mistaken as real and is meant to offer transparency around deep fakes.

Style isn’t explicitly protected under copyright law, but blocking such requests may be wise to avoid adding fuel to the litigation fire.

Authors who are angry about AI training should be glad that researchers are trying to demonstrate, with evidence, how market harm is occurring.

Last week, Substack rolled out AI detection site-wide for content, which has sparked considerable discussion, controversy, speculation, and confusion.

The subject is a retiree who has published more than 450 AI-generated books and sells a few hundred books—mostly print—per month.

Substack announced that readers can now use Pangram to scan text on Substack to see how much of it was likely written with AI assistance.

Service awards paid to class representatives (e.g., Andi Bartz) were also reduced from $50,000 to $15,000, for being “unreasonable”.

Google used access to digital copies of books via Google Books, the Google Play bookstore, and Google Scholar to train its Gemini model.

The study was performed by Edison Research, a leader in audio market research, and commissioned by Spoken, an AI audiobook production company.

A user discovered that writing copied and pasted from Claude into Archive of Our Own retains code that points to some level of AI involvement.

Publishing professionals have a responsibility to learn about AI—not to use it, but to effectively deal with its consequences in their work.

Scandals like this will continue unabated if AI use is prohibited and/or taboo but detection and enforcement is left in the hands of readers.

News Corp, parent company of HarperCollins and the Wall Street Journal, is negotiating deals with tech companies, seeking licensing income.

The Guild has consolidated all AI-related contract clauses—existing ones and new provisions—on a dedicated page on its website.

This latest AI lawsuit seeks class action status against Meta for pirating millions of books to train Meta’s LLM, Llama.

The guidelines address usage, disclosure, and tips on understanding and managing AI risks.

Their statement is in response to a report that some publishing professionals are feeding manuscripts into AI without explicit permission from authors.