The European Union’s draft artificial intelligence rules face a tight timetable. Lawmakers have yet to settle the terms, and an agreement before December appears unlikely. The negotiations cover both foundation models and high-risk AI systems.
Approval still depends on agreement
The draft must be approved by the European Parliament and EU member states. That means the negotiations need to produce terms acceptable to both before the rules can be finalized.
The process is taking place through trilogues. A fourth trilogue is upcoming, but a final agreement is not expected at that meeting. Another round is scheduled for early December, leaving little time to resolve outstanding disagreements before the year ends.
The remaining work is not just about reaching a date on the calendar. Lawmakers must also settle how the draft rules should apply to different kinds of AI systems, including foundation models. Until those points are resolved, the approval process remains unfinished.
A proposed threshold for foundation models
Spain, which holds the EU presidency, has put forward compromises intended to speed up the negotiations. One proposal would use a tiered approach to regulating foundation models with more than 45 million users.
A tiered system would distinguish between models according to the proposal’s criteria, rather than treating every foundation model in exactly the same way. The user threshold is one part of Spain’s suggested approach. It is a proposal in the negotiations, not a settled feature of the final rules.
That distinction is central to the debate. Supporters of a tiered approach may see it as a way to move talks forward, while opponents warn that smaller platforms could be just as risky. Their objection raises a basic challenge for the draft: how to set regulatory categories without overlooking systems that may pose risks despite having fewer users.
The dispute is about where to draw the line
The proposed threshold offers one possible dividing line, but it does not end the disagreement. If rules focus on models above a certain level of use, opponents argue, smaller platforms still need consideration. The negotiators therefore face a balance between Spain’s proposed compromise and concerns about how risk is assessed.
The draft also covers high-risk AI systems, alongside foundation models. That broad scope means the discussions include more than the question of which foundation models fall into a particular tier. The available account does not detail all of the disputed provisions, but it makes clear that lawmakers have not yet completed the overall agreement.
For the negotiations, the practical consequence is that a compromise on one proposal may not, by itself, settle the full draft. The European Parliament and EU member states still need an agreed text to approve. Until the parties reach that point, the rules remain under negotiation.
Early December is the next major opportunity
The fourth trilogue is not expected to produce a final agreement. The fifth trilogue, scheduled for early December, is the next identified opportunity to make progress. That timetable puts pressure on negotiators, but a scheduled meeting does not guarantee that they will resolve the disagreements.
If they fail to reach an agreement, negotiations could continue into next year, according to Reuters. That possibility follows directly from the unresolved issues and the limited time before the year-end deadline. For now, the central question is whether lawmakers can settle the approach to foundation models and complete the wider draft in time.
The proposed tiered system shows one effort to accelerate the process. The warning about smaller platforms shows why that effort remains contested. Until those concerns and the rest of the draft are addressed, the EU’s AI rules are still waiting on agreement.