AI Regulation ReferenceAI Basic Act
Obligations Covered
High-Impact AI Risk Management #
Requirements
| Requirement | Details |
|---|---|
| Risk identification | Operators must identify, assess, and mitigate risks for high-impact AI systems |
| Compute threshold | Systems trained with ≥10²⁶ floating-point operations classified as high-impact |
| High-risk sectors | 11 sectors including employment, healthcare, loans, criminal justice |
| Self-assessment | Must self-assess high-impact status pre-deployment and seek authority confirmation |
| Monitoring | Continuous monitoring of social impacts |
| Safety and reliability | Ensure safety, reliability, and human oversight |
Penalties
| Violation | Fine |
|---|---|
| Non-compliance | Administrative corrective actions; one-year guidance period until Jan 2027 |
AI Transparency and Disclosure #
Requirements
| Requirement | Details |
|---|---|
| Content watermarking | Mandatory watermarking/disclosure for AI-generated content |
| User notification | Users must be informed when AI systems make decisions affecting them |
| Documentation | Document operations, governance procedures, and content provenance |
| Age-appropriate protections | Additional disclosure requirements for minors |
Penalties
| Violation | Fine |
|---|---|
| Non-compliance | Administrative corrective actions |
AI Governance and Human Oversight #
Requirements
| Requirement | Details |
|---|---|
| Human oversight mechanisms | High-impact AI must include human oversight controls |
| Presidential Council | National AI Strategy Council oversees implementation |
| On-site inspections | Authority may conduct on-site inspections and data requests |
| Corrective measures | Authority may order corrective actions for non-compliant systems |
Penalties
| Violation | Fine |
|---|---|
| Non-compliance | Administrative corrective actions; on-site inspections |