APREX · EST. 2026
The EU's landmark artificial intelligence law is approaching a key milestone. In just a few days, on 2 August 2026, the transparency provisions of Article 50 of the EU AI Act will enter into force — and they apply not only to European companies.
What the law actually requires
The requirement has two parts. First, anyone offering or deploying AI systems that interact with humans must make it clear to the user that they are communicating with a machine. This disclosure must be given at the first interaction and must not be buried in the fine print.
Second, anyone using AI to generate or manipulate image, audio or video content that constitutes a deepfake must clearly state that the content has been artificially produced or altered. The labelling obligation applies even where there is no intent to deceive — what matters is whether an average person could reasonably perceive the content as genuine, according to the source material.
In addition, providers of generative AI models are required to make synthetic output machine-readable, for example through watermarking or metadata.
Norwegian relevance: the EEA Agreement binds Norway
Norway is not an EU member state, but through the EEA Agreement it is obliged to implement large parts of EU legislation. The EU AI Act is a regulation with EEA relevance, and Norwegian companies and developers should prepare for the requirements to apply in full in Norway as well. The transparency rules already apply to any Norwegian business offering services to users in the EU.
The labelling obligation applies even where there is no intent to deceive.
Deferred: the heavy requirements wait until 2027
It is important to distinguish between the different phases of the law. Although the chatbot and deepfake rules are now kicking in, the requirements targeting high-risk AI systems — such as tools used in recruitment, biometric surveillance and migration decisions — have been pushed back to December 2027 and August 2028, according to Silicon Canals.
This has caused confusion in the business community. A number of companies are reported to have interpreted news of the high-risk rules' postponement as meaning they have plenty of time on all fronts. That is a misreading.
Heavy fines and incomplete guidance
Breaches of the transparency requirements can trigger fines of up to €15 million, or 3 percent of the company's total global annual turnover — whichever is higher. For large technology companies, this could prove extremely costly.
One complication is that the EU's final guidance on Article 50 has not yet been published — only a draft. On 10 June 2026, independent experts presented a voluntary code of conduct for transparency in AI-generated content, but the underlying statutory obligation is not voluntary.
Technical challenges undermine the law's effectiveness
Experts have pointed out that watermarking as a method for identifying AI-generated content is vulnerable and can easily be removed, particularly by bad actors. Such actors will never voluntarily identify their own synthetic content, according to the source material.
Another contentious issue is that the law regulates how content is made, not whether it is harmful. This could impose new legal obligations even for harmless, accurate AI-generated content — while equivalent human manipulation remains unregulated.
Research also suggests that AI labels are not neutral: they can erode trust and perceived credibility, raising questions about whether labelling alone is an effective instrument.
What should Norwegian businesses do now?
Companies that use chatbots in customer communications or produce AI-generated image and video content aimed at European users should immediately review their solutions against Article 50. This includes ensuring that users are notified at their first point of contact with an AI, and that deepfake content is labelled consistently.
Given that final EU guidance has not yet been issued, industry experts recommend following the draft closely and monitoring updates from the European Commission in the months ahead.