The EU’s Article 50 is changing AI and it could affect your work
Article 50 of the EU’s AI Act (introduced 2 August 2026) has just changed the trajectory of AI use with immediate compliance measures being announced by global tech platforms that will make it possible to track AI generated content through invisible watermarks. While not yet compelled to be public facing, it marks a systemic change towards a technical separation of AI vs human content and signals a policy-led shift on how AI may be tracked and managed in the long-term.
It’s potentially good news for creators who are worried about their IP and the devaluation of their creative work in a sea of AI generation. It’s also a particularly welcome intervention in areas such as journalism, legal and policy documents, research papers, political campaigns etc… where problematic social impact has already surfaced as a result of AI use. But, at this early stage, the technical understanding of the Article 50 law – and it’s enforcement – is complex (and untested) with plenty of ambiguity. Especially if you’re creating in South Africa.
What is Article 50
While the application of the law is still surfacing, there are potential fines of up to €15 million or 3% of worldwide annual turnover on the line, so it’s worth getting up to speed on what this new legislation means. Article 50 is a European Union (EU) policy stipulation that requires digital content and services providers to inform users “when they are interacting with an AI system or where content is AI-generated.” (Artificial Intelligence Act (Regulation (EU), 2026).” While the legislation is currently for the EU, a globalised digital ecosystem, and the near certainty that many other countries will replicate some form of the EU’s policy framework, has meant that Article 50 has already triggered immediate changes to the way information is offered to digital users.
Global Tech Is Taking This Seriously
Big tech isn’t just responding to Article 50 in isolation, many of these platforms are also designated “Very Large Online Platforms” (VLOPs) or “Very Large Online Search Engines” (VLOSEs) under a related EU law, the Digital Services Act, which applies extra scrutiny to any service reaching more than 45 million monthly users in the EU.
Facebook, Instagram, YouTube, Google Search, Amazon Store, LinkedIn and TikTok all carry this designation and it sits on top of Article 50. VLOPs and VLOSEs face their own obligation to assess and mitigate the risks of AI-generated content spreading on their platforms, including making deepfakes clearly identifiable. Two overlapping regimes are now pointing at the same problem, which is a large part of why the response has been so fast and so visible. Google has announced that users will see labelling on AI image outputs and synthetic content via Google Lens, and Anthropic, the company behind Claude, has begun watermarking all of Claude’s AI-generated content. Users can’t see these watermarks directly yet, but Anthropic has promised to publish technical details on how they can be detected.
What Content Does it Affect
Platforms based anywhere in the world that are deploying AI into the EU will need to disclose where users are interacting with AI tools such as chatbots, virtual assistants and automated phone systems. Deployers of biometric and facial recognition technologies must also inform users when they are exposed to AI systems but defers to local “applicable laws”- most notably the EU’s General Data Protection Regulation (GDPR) for enforcement.
Deep Fakes – But Only For Fun
Where the legislation becomes interesting for creators is that deep fake disclosure, although now compulsory, is governed by use case. Publishers and content producers must disclose, “AI-generated or manipulated images, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful)” (Artificial Intelligence Act (Regulation (EU), 2026) but there is a caveat which imposes less stringent regulations on content, “where deepfake content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme”. Here, Article 50 states that publishers or content producers must still disclose their use of AI technology in media production, but “in an appropriate manner that does not hamper the display or enjoyment of the work” (Artificial Intelligence Act (Regulation (EU), 2026; The EU AI Act’s Transparency Rules: A Practical Guide to Article 50, 2026).
In practice, that means the disclosure sits around the work rather than on top of it. Think a credits line, a caption, or an opening note, rather than a watermark stamped across the image itself. The exemption is narrower than it sounds, though: it only protects work that’s genuinely artistic, not content that’s commercial or informative dressed up as creative, and it never excuses using a real person’s likeness in a misleading way.
Burden of Proof Lies with Humans
Much of what Article 50 proposes has grey areas and it’s likely that the regulations’ immediate goal is to regulate big tech – which seems to be working. It does however begin to set a precedent for how we think about AI generated content, how we keep receipts and where we incorporate it into our work.
One part of the regulation states that work is considered exempt where its “obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect” that AI technology is involved (Artificial Intelligence Act (Regulation (EU), 2026). In an international context, where unequal levels of digital literacy and heightened cultural diversity are present, these assumed ideas of what is “obvious” may be debatable and interpretation will need to be more refined as the law leads common practice.
What Does This Mean for South Africa
South Africa’s digital landscape differs immensely from the EU. The country currently lacks sufficient regulatory frameworks for digital communication, AI technologies and infrastructure to effectively implement Article 50’s proposed safeguards. Ambiguity surrounding how South Africans may comply with Article 50 is worsened by the fact that many digital platforms, particularly those owned by Global North entities, also lack specific regulations for South African users. For example, Meta simply advises South African advertisers to either comply with local regulations (which do not currently regulate digital advertising or AI use), or use their own judgement or discretion to ensure adherence to platform guidelines (Meta, 2025b). This lack of nuance is possibly because guidelines for advertisers in ‘other territories’ have been crudely adapted from those outlined for other regions and are fundamentally underdeveloped.
Although Meta has recently agreed to comply with Article 50 (Meta, 2026), their platforms’ content moderation infrastructures currently lack rigorous AI-detection frameworks. Content published on Meta platforms is only detected and flagged as ‘AI-generated’ if the creator/user produced the advertisement or post using Meta’s inbuilt AI suite.
After a retraction of South Africa’s original draft AI policy, after it was found to have fictitious AI references (the very thing these new laws seek to curtail), the renewed process of drafting South Africa’s AI Policy is currently underway with an independent panel of technology and legal experts led by Professor Benjamin Rosman of Wits University’s MIND Institute. That team will no doubt have a keen interest in the new EU laws and how they are enforced locally.
We’ll be keeping a keen eye on what this means for creators and how you can protect your work and stay compliant online.
By Sascha Polkey, BASA Insights & Advocacy
