Summary of an article published on The Web Scraping Club on 2024-05-28 (free to read). Read the full article

The EU AI Act is approved, and like GDPR it reaches companies outside the EU. If you offer AI services in the EU or to EU citizens, you are in scope wherever you are based. That covers companies that build AI systems and companies that use them in their work, so most businesses touching AI will have to check where they stand.

In this fourth Legal Zyte-geist column, Sanaea Daruwalla, Chief Legal & People Officer at Zyte, gives a practical overview of the regulation. She explains who must comply, when the rules start to apply and how the risk-based approach sorts AI systems into categories with different obligations. She then turns to the questions most companies ask first: what a chatbot like ChatGPT means for you, how much transparency is required, and what to put in place now. The usual disclaimer applies: this is information, not legal advice.

What’s inside

  • The Act separates providers, who develop AI systems, from deployers, who use them professionally, and both are covered.
  • Compliance is staggered: prohibited uses apply first, then generative AI provisions, then general compliance.
  • There are four risk categories, and obligations stack, so a higher tier also has to meet the duties of the tiers below it.
  • Chatbots are treated as Limited Risk, which brings a duty to disclose AI use and, often, an explainability statement covering data sources, training and decision logic.

In the full article: The risk-category chart, the list of conformity assessment obligations for High Risk systems and the preparation steps, from the AI register to audits and annual staff training, are only in the full article.

Read the full article on The Web Scraping Club