The European Parliament’s official position on the EU’s Artificial Intelligence Act (AI Act) prioritizes robust safeguards protecting people from harm; negotiations now enter the final round

(BERLIN, GERMANY | JUNE 14, 2023) — Today, the European Parliament voted to adopt its position on the EU’s landmark AI law, the AI Act. This marks a major milestone in the legislative process: policymakers are now gearing up for the final round of negotiations between the European institutions.

It also marks another important milestone: strong AI regulation is on the horizon at a time when the technology is becoming even more capable and pervasive.

As Mozilla has stated before, the AI Act has the potential to make AI across the EU more trustworthy — And the European Parliament vote today marks further progress toward robust safeguards against AI harms. In particular, the proposal adopted by the European Parliament includes several improvements that Mozilla has actively pushed for: more due diligence and accountability along the AI supply chain to account for generative AI and other general-purpose AI models; more transparency around AI; and new rights for individuals to push back against AI harms.

Says Mark Surman, President of the Mozilla Foundation: “The European Parliament has fought hard for new protections and safeguards against the harms of AI. This new proposal would make AI developers more accountable and create more transparency around AI systems — including ones like ChatGPT. Now, it’s important they keep up the fight to ensure that these accomplishments aren’t lost on the final stretch of negotiations.”

Specifically, the European Parliament proposal includes the following changes in areas of key importance to Mozilla’s work on trustworthy AI:

• General-purpose AI/”foundation models”: As opposed to the original proposal from the European Commission, the European Parliament proposal includes upstream due diligence obligations for developers of so-called “foundation models” or generative AI, in order to better prevent and mitigate harms further down the AI value chain.
  
  
• Public database for (high-risk) AI systems: The obligation to register AI systems in a public EU database was extended to public-sector deployers of high-risk AI systems (as opposed to only developers) in order to provide more transparency about the specific context in which high-risk AI systems are used by public authorities. Other deployers will also be able to register AI systems they put to use voluntarily.
  
  
• Complaint mechanisms: Individuals or groups affected by AI systems are equipped with the right to file complaints with supervisory authorities to push back against potential harms.

Still, now is not the time to sit back — neither for lawmakers, nor for civil society. During trilogue negotiations between the European Parliament, the European Commission, and EU member states, crucial issues will be on the table again. This also includes shortcomings of the European Parliament’s position, like the weakening of the classification process for what categories of AI systems are considered high-risk.

At Mozilla, we remain committed to the values we placed at the center of our advocacy on the AI Act over a year ago: ensuring accountability, creating systemic transparency, and giving individuals and communities a strong voice. We call on the European institutions to do the same — these values are not optional, but non-negotiable underpinnings of a strong AI Act that puts people first. As an independent and trusted voice in the tech sector, Mozilla stands ready to work with EU lawmakers to make sure of this.