AI Act
·
11 months ago
The European Commission published draft Guidelines on 18 July 2025 clarifying how the EU AI Act applies to General Purpose AI models, establishing a compute threshold of 10²³ FLOPs for GPAI classification and 10²⁵ FLOPs for systemic-risk designation. Providers must maintain documentation, publish training data summaries, comply with copyright policies, and those exceeding the 10²⁵ FLOPs threshold must notify the Commission within two weeks and undergo comprehensive risk assessments. The guidelines establish enforcement timelines with the AI Office gaining full enforcement powers on 2 August 2026, while providers of models placed before that date have until 2 August 2027 to achieve compliance.
AI Act
·
11 months ago
The EU AI Act's Code of Practice provides a framework for developers of General Purpose AI models to comply with regulatory requirements, with enforcement beginning August 2, 2026 for models released after August 2, 2025, and a compliance deadline of August 2, 2027 for existing models. Key commitments include maintaining comprehensive documentation for at least 10 years, establishing copyright policies that respect machine-readable rights signals like robots.txt, implementing safety and security frameworks with risk assessment and incident reporting within 2 days for critical infrastructure impacts, and ensuring transparency with downstream providers receiving documentation within 14 days. Signatories must assign clear organizational accountability for oversight and retain detailed safety records for a minimum of 10 years, with public disclosure of safety frameworks recommended unless models meet specific safer criteria.
