On 2 August 2025 the EU AI Act's obligations for general-purpose AI models became applicable: technical documentation, copyright policies, training-data summaries, and for the most capable models, systemic-risk assessments. The governance machinery, the AI Office and national authorities, is now running, and the countdown to the high-risk system obligations of August 2026 has begun.
Downstream, the problem is provenance
For businesses that use rather than build these models, the practical 2025 problem is different: as generated text, images, audio and video become indistinguishable from captured reality, how does anyone prove a piece of content is authentic, unaltered, and from the claimed source? Watermarking helps at the model level but does not survive screenshots, re-encoding or adversaries. The Act's Article 50 transparency duties (disclosing AI-generated content) need an enforcement-grade counterpart.
Cryptographic provenance exists today
That counterpart is the qualified electronic seal: a cryptographic attestation, backed by an EU Qualified Trust Service Provider, that this exact content existed in this exact form at this time from this source. WeVerify's Qualified e-Seal covers documents, data streams, video recordings and AI outputs €0.01 per seal, and maps directly to the AI Act's data-provenance expectations. In a world where anything can be generated, the ability to prove what was not becomes the asset.