EU AI Act
The first comprehensive dedicated law globally for regulating AI, adopted by the European Union in 2024. It employs risk-tiered regulation, imposes transparency obligations on general-purpose AI and foundation models, and provides for fines of up to 7% of global revenue.
Core Facts
The EU AI Act classifies medical auxiliary diagnosis systems as high-risk AI systems and requires verifiable decision-making rationale to be provided.
Penalties for violations of the EU AI Act can reach up to 7% of global revenue or €35 million.
The EU AI Act requires machine-readable labeling of deepfake/synthetic content.
Timeline (last 90 days)
The COMPL-AI project provides a technical interpretation of the EU AI Act, organizing compliance evidence into six principle groups and eighteen technical requirements, twelve of which can be measured through existing benchmarks.
This is the first major enforcement action since the EU AI Act took effect; if major safety lapses are found, the companies could face fines of up to 3% of global annual turnover.
The EU AI Act adopts differentiated granularity based on risk levels: unacceptable risk is directly prohibited, high-risk scenarios are subject to strict requirements, and low-risk scenarios are only required to meet transparency obligations.
The EU AI Act sets specific restrictions on emotion recognition and biometric categorization; several US states require devices involving recording to fulfill two-party consent obligations.
The EU AI Act requires disclosure of training data summaries, but does not directly resolve the issue of copyright licensing.
The EU AI Act has classified deepfake voices as a category requiring explicit disclosure.
The EU AI Act mainly relies on administrative penalties, with fines of up to 7% of global revenue, and does not directly authorize criminal prosecution, and whether data scraping itself is illegal is still left to each member state's copyright law.
The EU AI Act requires transparency labeling for AI-generated content, but does not make explicit provisions regarding data attribution for watermarking technology.
The EU AI Act uses a training compute threshold of 10^25 FLOP as one of the benchmarks for determining frontier models.
Article 12 of the EU AI Act requires that event data automatically captured by the system includes activation and deactivation records, input data characteristics, and output results.
13 more timeline events
All Facts (20)
The EU AI Act classifies medical auxiliary diagnosis systems as high-risk AI systems and requires verifiable decision-making rationale to be provided.
85%VerifiedPenalties for violations of the EU AI Act can reach up to 7% of global revenue or €35 million.
80%VerifiedThe EU AI Act requires machine-readable labeling of deepfake/synthetic content.
65%VerifiedBoth the EU AI Act and China's Interim Measures for the Management of Generative Artificial Intelligence Services impose compliance requirements on data processing activities by enterprises using AI tools.
65%UnverifiedAI安全领域正从整体基准测试的一刀切评估转向领域特异性的精细化治理趋势
70%UnverifiedThe EU Artificial Intelligence Act of 2024 formally requires that AI-generated content must carry machine-readable labels.
60%UnverifiedThe EU Artificial Intelligence Act was formally passed in 2024, classifying real-time remote biometric identification systems in public spaces as high-risk or even unacceptable-risk categories.
60%UnverifiedThe COMPL-AI project provides a technical interpretation of the EU AI Act, organizing compliance evidence into six principle groups and eighteen technical requirements, twelve of which can be measured through existing benchmarks.
50%UnverifiedThis is the first major enforcement action since the EU AI Act took effect; if major safety lapses are found, the companies could face fines of up to 3% of global annual turnover.
50%UnverifiedThe EU AI Act adopts differentiated granularity based on risk levels: unacceptable risk is directly prohibited, high-risk scenarios are subject to strict requirements, and low-risk scenarios are only required to meet transparency obligations.
50%UnverifiedThe EU AI Act sets specific restrictions on emotion recognition and biometric categorization; several US states require devices involving recording to fulfill two-party consent obligations.
50%UnverifiedThe EU AI Act requires disclosure of training data summaries, but does not directly resolve the issue of copyright licensing.
50%UnverifiedThe EU AI Act has classified deepfake voices as a category requiring explicit disclosure.
50%UnverifiedThe EU AI Act mainly relies on administrative penalties, with fines of up to 7% of global revenue, and does not directly authorize criminal prosecution, and whether data scraping itself is illegal is still left to each member state's copyright law.
50%UnverifiedThe EU AI Act requires transparency labeling for AI-generated content, but does not make explicit provisions regarding data attribution for watermarking technology.
50%UnverifiedThe EU AI Act uses a training compute threshold of 10^25 FLOP as one of the benchmarks for determining frontier models.
50%UnverifiedArticle 12 of the EU AI Act requires that event data automatically captured by the system includes activation and deactivation records, input data characteristics, and output results.
50%UnverifiedCompanies that violate the EU AI Act can face fines of up to 3% of their global turnover.
50%UnverifiedThe EU AI Act explicitly prohibits most real-time public biometric identification scenarios.
50%UnverifiedDuring the legislative process of the EU AI Act, safety researchers, open-source advocates, and commercial interests each held their own views, causing the regulatory framework to be revised multiple times.
50%