Seems like a feature. This was always going to be the case. Just like how it was cheaper to train those models on billions of prior works than to have generated or paid to generate all those works in-house.

> "One test for potential investors will be their willingness to support Sovereign Australia AI’s decision to earmark $10 million of its future funding to compensate copyright owners for the data used to train its model. This includes working with news services under a paid model and buying books and music where needed."

> “We don’t want the adversarial relationship of most other AI builders around the world who chose not to take that proactive approach to copyright.”

> "Sovereign Australia AI said it would not scrape the pages of publishers who have added “robot.txt” files to their web pages. This is a line of code that tells bots not to scrape the information, but it is frequently ignored. The company will add a meta tag to every piece of data it acquires, recording where it came from and how it was sourced."

> To build its model, Sovereign Australia AI says it has placed Australia’s largest-ever order for sovereign AI capacity: 256 of the latest Nvidia Blackwell B200 GPUs which will be hosted inside one of NextDC’s Melbourne data centres

So... almost the exact opposite. Please read the article before commenting next time.