The law also sets up an AI Office within the Commission, which will work in conjunction with a new scientific advisory panel to aid the body in understanding the capabilities of advanced AI models for the purpose of regulating systemic risk. The law will likely not be in full force until 2026, but the EU will be pushing for AI models to abide by codes of practice in the meantime. The law also includes a list of strictly prohibited use-cases of AI, which will apply six months after the law enters into force.
Key takeaways:
- The European Union lawmakers have agreed on a comprehensive law for regulating artificial intelligence, which includes powers for the Commission to adapt the pan-EU AI rulebook to keep pace with developments in the field.
- The law includes a low risk tier and a high risk tier for regulating general purpose AIs (GPAIs), with the high risk rules applying to generative AI technologies determined by an initial threshold set out in the law.
- The EU AI Act references the amount of compute used to train the models, setting the bar for a GPAI to be considered to have “high impact capabilities” at 10^25 FLOPs.
- A new expert oversight body will be set up within the Commission, called the AI Office, which will develop other benchmarks over time and play a key role in setting risk classification thresholds for GPAIs.