The EU’s AI law comes online

The Future. The European Union’s A.I. Act, which focuses on the riskiest aspects of AI, is finally drafted and agreed on after nearly three years — although there’s still the fine print to work out and a final vote for official passage. But with most of the policies being inactionable until 2025 and global enforcement still a question mark, there may be a possibility that the legislation could become outdated or toothless without aggressive implementation.

Guardrail generation
The A.I. Act is the first comprehensive regulatory legislation in the industry with unique protections.

• AI-generated images and deepfakes will have to be labeled as such, putting the responsibility on companies — not customers.
• The biggest AI companies will have to provide detailed summaries of their training data and provide risk reports.
• Bulk facial-image scraping is now a no-no, as is emotion-recognition systems when employed in the workplace or educational settings.
• Social-scoring systems that measure a person’s loyalty are off the table.
• Facial recognition tech in law enforcement will be limited, with exemptions for identifying victims in cases of terrorism, human trafficking, and kidnapping.

Firms in violation of the rules could be fined up to 7% of their global sales.

While the EU is applauding the comprehensive agreement among its 27 member countries as a foundational document for global legislation, it’s also facing pressure from NGOs, governments, and companies on either side of the regulation debate, meaning consensus is still… well… up for debate.