EU AI Act Compliance for General-Purpose AI Models Begins August 2, 2025, Systemic Risk Defined by 10^23 FLOPs

Brussels – The European Union's AI Act has officially commenced its obligations for providers of general-purpose artificial intelligence (GPAI) models as of August 2, 2025, ushering in new layers of compliance for the rapidly evolving technology sector. This significant regulatory step aims to enhance transparency, safety, and accountability within AI systems operating across the EU market. The move follows the European Commission's release of detailed guidelines on July 18, 2025, clarifying the scope and requirements for AI model providers.

The new framework mandates that GPAI model providers adhere to specific rules, including those related to copyright law and the public availability of detailed summaries about training data. As noted by HackerNoon on social media, the regulations introduce "layers of compliance for general-purpose AI models," emphasizing the comprehensive nature of the new legal landscape. The guidelines are complemented by a voluntary General-Purpose AI Code of Practice, published on July 10, 2025, designed to assist developers in meeting these new requirements.

A key aspect of the guidelines involves defining systemic risk for highly capable AI models. Models exceeding a computational threshold of 10^23 FLOPs (floating point operations) are now classified as posing systemic risks, necessitating thorough evaluations and reporting of serious incidents to the European Commission. This threshold provides a concrete metric for identifying the most impactful AI systems requiring stringent oversight.

The regulations also address the complex area of open-source AI models. While certain open-source models may be exempt from some obligations, this exemption does not apply to those deemed to carry systemic risks. This nuanced approach, described in the tweet as concerning "open-source loopholes," ensures that even freely available models are subject to scrutiny if their capabilities could lead to widespread impact or harm.

Providers of GPAI models placed on the market after August 2, 2025, are immediately subject to these new compliance requirements. For models introduced before this date, a grace period extends until August 2, 2027, to achieve full compliance. The European Commission’s enforcement powers, including the imposition of fines for non-compliance, will become fully applicable from August 2, 2026, marking a critical deadline for the industry.