Since 2 August 2026, Article 50 of the EU AI Act has been in force across the European Union, requiring warnings on deepfakes. Yet for anyone uploading an AI-generated clip to TikTok or YouTube, the first rule that bites is not the law but the platform’s own policy, which has demanded disclosure for some time and reserves the right to pull content that skips it.
Platforms label on their own what they can detect, especially when a file carries C2PA metadata or was produced with their in-house AI tools. Users must flag content themselves when it looks real and was generated or substantially altered with AI. Failure to do so risks removal or lost monetisation. European law, meanwhile, targets those using AI in a professional capacity, not the casual poster sharing for fun.
- Automatic labelling: YouTube, TikTok and Meta apply labels when they detect C2PA credentials or content made with their own AI tools.
- Creator disclosure: Realistic synthetic video or audio must be flagged manually at upload, and some labels cannot be removed once applied.
- Legal deadline: Article 50 applies from 2 August 2026, with a partial extension to 2 December 2026 for providers already selling systems.
When the platform labels content for you
The trigger is a signal the platform can read, or detection by its own systems. The most widespread signal is the Content Credentials standard known as C2PA, metadata that some generation tools attach to images and video.
YouTube states in its policy on AI-generated content, in its November 2026 version, that it may apply the label automatically in three cases: content made with its generative AI tools, content carrying C2PA metadata, and content its internal systems detect as generated or altered with AI. If the system gets it wrong, creators can correct it in most cases, but not when the label stems from C2PA, YouTube’s own tools or a manual review.
TikTok describes in its help centre, consulted in November 2026, an automatic “AI-generated” label for content made with its AI effects or arriving with C2PA Content Credentials. Once applied, it cannot be removed.
Meta, covering Instagram, Facebook and Threads, announced in February 2024 that it would read AI indicators from the C2PA and IPTC standards to label images made with tools from Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock. In updates during July and September 2024 it renamed the label from “Made with AI” to “AI info” and moved it into the post menu when AI was used only for editing or retouching. The visible label was kept for content detected as generated.
When disclosure falls to the creator
The obligation applies when content looks real and was generated or significantly changed with AI.
On YouTube, disclosure happens at upload: in YouTube Studio, under Attributes, the “AI use” option, selecting Yes. The categories requiring a flag include a real person appearing to say or do something they did not, altered footage of real events or places, realistic scenes that never happened, and generated music used as a central element. Exempt are unrealistic content, beauty filters, colour adjustments, subtitles, AI-assisted scripts or thumbnails, and cloning your own voice for dubbing.
On TikTok, the “AI-generated content” toggle is activated before publishing from More options. Realistic AI-generated images, audio or video are mandatory; significant edits such as people doing or saying what they did not, or unrecognisable faces after a face swap, are recommended. Content made solely with TikTok’s own AI effects is labelled automatically unless further altered with another AI tool.
On Meta, photorealistic video or realistic audio created or digitally altered must be disclosed. AI retouching is exempt from the visible label: since September 2024 that disclosure sits in the post menu rather than on display. Meta announced a disclosure tool at publishing in 2024, though no current help page with the exact menu name for Instagram and Facebook was found.
Penalties, prohibitions and what Article 50 actually demands
Consequences vary by platform. YouTube warns that creators who systematically fail to disclose may have the label applied manually, content removed or access to the Partner Programme suspended, while stressing that declaring AI does not limit reach or monetisation. TikTok may remove unlabelled content that breaks its rules, and labelling content that is not altered also breaches its terms. Meta said in 2024 it could penalise those who skip the disclosure tool when required.
A label is no safe harbour. TikTok bans, even when labelled, AI content featuring false authorities or crises, public figures endorsing products, or the likeness of minors and private adults without permission.
Article 50 imposes two distinct duties: providers of generators must mark outputs in a machine-readable format, and professional deployers must disclose deepfakes. The Regulation (EU) 2024/1689 sets out in Article 50.2 that providers of systems generating synthetic audio, image, video or text must mark results in a machine-readable format allowing detection as artificial. Article 50.4 obliges deployers of deepfake-generating systems to make public that the content is artificial; for clearly creative, satirical or fictional works, a disclosure that does not spoil the work suffices.
For ordinary users, the definitions matter most. Article 3 defines a deployer as anyone using an AI system under their authority, except when use falls within a personal, non-professional activity. The reading here is that the legal duty lands on companies, media outlets, agencies and creators working with their content, not on someone posting a montage to a personal profile. That user remains bound by each platform’s rules.
The Regulation (EU) 2026/1744, the digital omnibus on AI published on 24 July 2026, kept 2 August 2026 as the general date but gave providers already selling systems before then until 2 December 2026 to comply with Article 50.2 marking.
The current system works well for those who want to disclose and poorly for those who want to deceive: automatic labels depend on metadata that, as Meta acknowledged in 2024, can be stripped. The real value of the European rule lies less in the badge you see on TikTok than in forcing manufacturers to mark everything they generate. Until that marking is robust, a label tells you what someone chose to declare, not everything that is artificial.