EU AI Act: Artificial Intelligence Under Supervision from August 2025

The European Union's AI Act will see its next phase of implementation on August 2, 2025, bringing further obligations for companies and authorities. This update particularly affects providers and users of General-Purpose AI (GPAI) models, such as GPT and LLaMA, introducing new requirements for documentation, transparency, risk management, and compliance.

Providers of GPAI models must adhere to new regulations demanding technical documentation detailing the model's architecture, training data, and computational power. They are also required to publish transparency reports outlining known weaknesses and risks and conduct tests to prevent discriminatory or harmful outputs. The sources of training data must also be disclosed in an understandable manner.

Models classified as systemically risky will face additional requirements, including mandatory risk assessments and adversarial testing to prevent misuse. EU member states must also designate national supervisory authorities and establish enforcement infrastructure.

Users of GPAI models will also encounter new obligations, especially when integrating these models into high-risk applications in sectors like HR, healthcare, or finance. Users must ensure that the deployed systems comply with the AI Act, including documentation and risk assessment duties.

Non-compliance can result in significant fines, potentially reaching up to €35 million or 7% of a company's global annual revenue. Given that many technical requirements are not yet fully standardized, companies should remain vigilant and prepare for these upcoming regulations.