US urges G-20 countries to allow AI training on creators' work

The United States is urging G-20 member countries to allow AI companies to train models on creators' work, according to a report from The Straits Times. The push, if it gains traction, could shape how major economies write copyright rules for generative AI training data.

At the heart of the dispute is the boundary between training data and copyright. Mainstream AI models depend on vast corpora of text and images, much of it produced and published by creators, which makes the question of whether, and how, such work may be used in training a central regulatory battleground.

As framed in the report, Washington's stance leans permissive: it wants national legal frameworks to leave room for training on creators' work rather than generally requiring case-by-case authorization up front. That position sits in tension with creator groups and some governments that demand prior consent and compensation.

The issue has risen to the G-20 level partly because training-data rules are inherently cross-border. Models are trained on material drawn from around the world, and the compliance risks that follow travel to every market where a model is deployed, which means no single country can settle the boundary on its own.

For creators, the outcome of the G-20 debate determines whether their work can be used in training with knowledge, authorization, and reward; for AI developers, it determines the availability of training corpora and the cost of compliance. The contest between these two forces will define the rulebook of the generative-AI content economy.

So far, public information is largely limited to the news report itself; which G-20 forum the United States raised the issue in, and the exact wording it proposed, are still emerging. Watch for the final language of G-20 statements, and whether major economies such as the European Union adjust their training-data copyright positions in response.