Summary of an article published on The Web Scraping Club on 2024-01-09 (free to read). Read the full article
Last year brought a big increase in lawsuits and regulation touching web scraping, mostly around Generative AI. If you scrape data to train an LLM, the legality of those inputs is what matters. Copyright and personal data are the two open questions, and no court or regulator has fully answered either yet.
We go through the 2023 picture and what to expect in 2024. On the US side, we separate the pure scraping cases from the GenAI class actions against OpenAI, Microsoft, Meta, StabilityAI and Google, plus the New York Times case against OpenAI. On the EU side, we cover the data protection authorities looking at OpenAI’s use of personal data under GDPR, and the provisional agreement on the EU AI Act with its risk-based approach.
What’s inside
- Why the outputs of GenAI systems matter less to scrapers than the inputs used for training.
- The argument from big tech that a machine learning from copyrighted data is no different from a human doing it, and why it is still untested in court.
- How US state exceptions for public personal data sit against GDPR’s lawful basis, notification and data subject access requirements.
- Open GDPR questions for models already built: whether personal data must be removed or training started from scratch.
In the full article: The full list of scraping cases to watch with the causes of action at play, the AI Act risk tiers and their requirements, and links to the StabilityAI and Meta ruling posts and Zyte’s compliance resources.