From today, AI content must be labelled – on two continents at once
The EU AI Act and California's SB 942 take effect the same day, requiring labelling of AI-generated image, video, audio and text. But the technology behind it has a fundamental weakness.
As of today, August 2, 2026, labelling AI-generated content becomes legally required on two continents at once. In the EU, the transparency requirements in the AI Act’s Article 50 take effect, and in California the law SB 942 becomes operative. The dates coinciding is no accident: California deliberately delayed its start to align with the EU.
Together, the day marks a turning point: the requirement to be able to know when an image, a video, a sound or a text was created by artificial intelligence moves from courtesy to law.
What the EU rules require
Article 50 of the AI Act imposes two main duties. Providers of AI systems – such as OpenAI, Google and Adobe – must embed a machine-readable mark in content their systems create, so it can be detected automatically. Those who use AI must in turn disclose it: deepfakes and realistic AI content must be labelled, AI-generated text on matters of public interest must be labelled, and users must be told when they are talking to a chatbot rather than a human.
The rules apply immediately, including to AI systems already on the market – there is no transition period for existing services. Fines can reach 15 million euros or 3 percent of global turnover.
California’s version: invisible authenticity stamps
The American law SB 942 targets large AI providers with more than one million users in California. They must build an invisible, machine-readable "authenticity stamp" into every image, video and audio file their system creates – with the name of the provider and system, a timestamp and a unique identifier. They must also offer a free tool where anyone can check whether a file is AI-generated. Violations are punished with 5,000 dollars per instance, per day.
Both frameworks rest in practice on the same technical standard, called C2PA – an industry standard for tracing the origin of digital content.
The technology is not watertight
Experts are quick to point out the limitations. An invisible authenticity stamp that sits in a file’s metadata can be removed by taking a screenshot or uploading the content to a platform that rewrites the file. More robust methods that embed the mark in the image pixels themselves are harder to remove, but do not survive all processing. And the absence of a mark proves nothing – it may mean the content is genuine, or that the mark was removed.
The marking thus reliably says "this comes from system X" when the stamp is intact, but cannot guarantee that something was not made by AI.
What does it mean for Norwegians?
The EU’s AI Act is EEA-relevant and is being incorporated into Norwegian law, and applies in any case to businesses targeting the EU market. For ordinary users, today’s milestone means AI content should gradually become easier to recognise – on social media, in advertising and in the news feed. For businesses and content creators, it means a new duty to be open about when the machine has been involved.
