Report says Anthropic trained AI on Korean authors' works without permission

A report by the Korean outlet Chosunbiz says Anthropic trained its artificial intelligence models on works by Korean authors without obtaining permission. The report frames the issue around licensing, stating that the works were used in model training without the authors' consent.

Based on the information available in the candidate item, the story is a specific allegation rather than broad industry commentary: the material was used for training, and authorization was missing. The headline and summary do not name the authors involved, the number of works, or whether any licensing discussions took place, so the full scale of the case remains unclear.

Training-data copyright has been one of the most persistent disputes in the AI industry. Large models are built on enormous collections of text drawn from public web pages, e-books, scanned books and published material, much of which remains under copyright, while developers rarely disclose their sources book by book.

For Anthropic, the allegation touches the least transparent part of model development: where training data comes from, how it was obtained, and whether any traceable licensing record exists. If authors ask for an explanation, the company would need something more specific than a broad fair-use argument.

Korea is not a marginal market for Anthropic. Local developers, startups and large enterprises work with overseas model services, and the country's authors and publishers are attentive to where their work ends up. A report like this can therefore reach both commercial customers and creators at the same time.

At the industry level, the episode is another sign that training-data governance is moving from a technical footnote to a compliance question. Beyond raw model capability, being able to explain data provenance and reach licensing arrangements is becoming part of what AI companies must demonstrate when they operate in different jurisdictions.

Three things are worth watching: whether Anthropic responds to the report, whether Korean authors and publishing groups take collective action, and whether local regulators fold training-data licensing into a clearer rule set.