EU AI Act: Obligations for General-Purpose AI Models Now Apply

On August 2, 2025, the EU Artificial Intelligence Act's obligations for general-purpose AI (GPAI) models began to apply: model providers must compile technical documentation, maintain copyright-compliance policies and publish summaries of training content; models with systemic risk face additional evaluation and reporting duties. An accompanying Code of Practice offers model vendors a compliance pathway.

For the language services industry, this wave of obligations transmits along two paths. One is direct: labeling duties for AI-generated content enter implementation preparation, and machine translation and AI writing services need technical readiness for "content identifiable as AI-generated". The other is indirect: greater transparency from model vendors (public disclosure of training data, capability boundaries and risks) gives language providers a fuller information base when selecting and evaluating models.

The EU compliance timeline is advancing steadily: in August 2026, the bulk of the Act's provisions will apply in full. Enterprises serving the European market — and their language providers — need to fold AI compliance into daily operations. This is no longer forward planning but a present requirement.

From the first bans to GPAI obligations, the EU AI Act's phased implementation offers the world an observable sample of AI governance. The language industry stands inside it — both as a subject of compliance and as a provider of compliance services.