How Expensive is the 'Compliance Cost' of Training Data for Large Models?

Anyone working in AI knows that training data is the dirtiest, most grueling part of the entire pipeline. Us folks building small SaaS tools usually scrape some public data for training, and we even worry about copyright issues when using APIs. Now, Anthropic's $1.5 billion settlement serves as a hard-hitting wake-up call for the whole industry.