Other

The EU AI Act's synthetic-content labelling rules are now in force — and they raise the premium on data that is verifiably real

Source: AIAct.eu · Nov 2, 2026

A significant piece of the EU AI Act took effect on 2 November 2026: AI-generated image, video and audio offered in the EU must now be marked as synthetic in a machine-readable way, and providers must ship a mechanism that lets that content be detected as machine-made. Systems already deployed get a brief window to fall in line, and the enforcement teeth are real — penalties scale into the millions or a share of global turnover. The practical effect is a bright line through the content supply chain: material that is synthetic now has to announce itself.

Most of the commentary will focus on the obligation side — watermarking, detection, the compliance lift for generative-media vendors. The more durable consequence is what it does to the value of the other side of the line. When synthetic content must be labelled and made detectable, content whose human origin can actually be demonstrated becomes the scarce, low-risk input. Authenticity stops being a vague virtue and becomes a property you either can or cannot prove.

For teams training and evaluating models, that reframes what 'good data' means. A dataset assembled from scraped or synthetic material now carries labelling and detectability questions with it. A dataset of real human interaction — captured first-hand, with the origin of every asset documented — carries the opposite: a clear, defensible answer to 'where did this come from, and is it real?'

That is the ground a consented interview and assessment corpus already stands on. Real people, recorded with consent, de-identified, and delivered with per-asset provenance, is as far from unlabelled synthetic media as training data gets. As regulation makes the synthetic path louder and more expensive to walk, the quiet advantage accrues to data that was authentically human from the first recording — and can show it.

Key Points

• As of 2 November 2026, AI-generated image, video and audio placed on the EU market must carry machine-readable markings identifying it as synthetic, backed by a detection mechanism
• Systems already in the field have a short grace period to comply; penalties run to the higher of substantial fixed fines or a percentage of global turnover
• The rule forces a hard line between content that is synthetic-and-must-say-so and content whose human origin can be demonstrated
• Data whose authenticity and origin are documented becomes the low-risk input precisely as synthetic media gets labelled and policed