The European Union’s landmark legal framework to regulate artificial intelligence (AI) received political agreement Friday, moving one step closer to official adoption.
The deal between the European Parliament and the Council on the AI Act advances the legislation to formal approval, which would put it on track to take full effect as soon as 2026. Certain provisions of the act, such as regarding general purpose AI, would apply within the first year of its adoption and entry into force.
During the transition period, the European Commission said in a press release it would launch an “AI Pact” that companies can voluntarily commit to to begin early implementation of the AI Act’s measures. European Commission President Ursula von der Leyen said in a statement around 100 companies have expressed interest in the pact.
“Until the act will be fully applicable, we will support businesses and developers to anticipate the new rules,” she said.
The AI Act follows a risk-based approach and regulates AI systems in accordance with the level of risk they present. There are four bands:
- “Minimal” risks, where the risk of harm is so low such systems are not regulated.
- “Limited” risks, which are subject to certain transparency requirements but should pose no serious levels of harm.
- “High” risks, which could include those that evaluate consumer creditworthiness, assist with recruiting or managing employees, or use biometric identification.
- “Unacceptable” risks, which are so potentially harmful they are banned from use.
Companies breaching the rules could face fines of up to 7 percent of global turnover or 35 million euros (U.S. $38 million), whichever is the higher figure.
Von der Leyen said the European Union intends for the act to “make a substantial contribution to the development of global guardrails for trustworthy AI.” She said the European Union would continue to work with its international partners regarding AI oversight.