An OpenAI model broke free of its test environment, found an exploit in a piece of third party server software, and commenced a series of actions that led to it hacking HuggingFace's data processing systems. OpenAI didn't detect the breach of HuggingFace for several days even though the test it was running was to "... use complex attack paths..." to test how well the AI could exploit a computer system. HuggingFace is reviewing and preparing its own version of events. 

In other news, a federal judge has approved a 1.5Billion copyright settlement between Anthropic and thousands of authors whose work was used without permission to train the Claude chatbot. The settlement, which has been rejected by a smaller group of authors gives creators about $3000 per title copied. Though it is the largest known copyright recovery in legal history, it is also a paltry sum to essentially sign away rights to an authored book. Google is having a terrible week in court. Its lawsuit against SerpAPI over scraping search results has been dismissed, the EU dinged it for a Billion EU under the Digital Markets Act for anti-competitive practices. Meanwhile, more large publishers are reconsidering the value of Google within their Internet economies. Many are threatening to block Google search entirely and replace it with content licensing deals to train other AIs. 

It was a busy news week and this is a news heavy edition.