The Act also bans certain AI applications and activities, such as scraping facial images from CCTV cameras and creating "social scoring systems". However, law enforcement agencies are exempt from these rules. The Act also provides for fines for companies found to have violated the rules, ranging from 35 million euros to 7% of their global revenue. The Act is unlikely to come into force before 2025.
Key takeaways:
- The European Union has reached a provisional agreement on the new EU AI Act, which is believed to be the world's first major regulation governing AI.
- The act categorizes AI into various levels of risk, from "unacceptable" to "low-risk", and requires AI companies to disclose information about how their models work and evaluate them for "systemic risk".
- There will be obligations for creators of "high-impact" AI systems, including risk assessments, adversarial testing, incident reports, and providing detailed summaries of the data used to train them.
- The act also bans certain applications and activities, such as scraping facial images from CCTV cameras, categorizing individuals based on sensitive characteristics, and creating "social scoring systems". Law enforcement agencies, however, are exempted from some of these rules.