Commentary. Europe’s AI rulebook has reached a consequential stage: transparency obligations under the EU AI Act became applicable on 2 August 2026, including requirements concerning certain AI-generated or manipulated content.[1] The European Commission also issued guidance intended to help providers and deployers identify, label and disclose synthetic material in practice.[2]
Our view is straightforward: this is necessary, but not nearly sufficient. A visible label can help people understand what they are seeing. It cannot, by itself, tell them whether the material is accurate, lawful, fairly made or responsibly distributed. Policymakers and technology companies should treat disclosure as the beginning of accountability—not its conclusion.
Why labeling matters
There is a strong case for requiring platforms and creators to disclose synthetic media. Images, audio and video can now be generated or altered at a speed and scale that outstrip ordinary editorial verification. A clear indication that content was produced or materially changed by an AI system gives audiences an important piece of context, particularly during elections, emergencies and breaking news.
Transparency can also protect legitimate uses. A filmmaker, educator or satirist should not be forced to defend an otherwise lawful work as though it were a covert deception. Clear provenance rules could allow creative and productive applications to flourish while giving audiences a better chance to distinguish simulation from documentation.
The EU’s approach is more useful than a vague appeal to “responsible AI” because it creates concrete obligations and assigns duties across the AI value chain. The legislation’s broader framework also differentiates among prohibited, high-risk and transparency-related requirements, with some high-risk obligations taking effect later.[1]
The limits of a sticker
Yet a label can become a fig leaf. People do not encounter media in a vacuum: they see it through recommendation systems, private messaging groups and political campaigns that may strip away context. A disclosure buried in a caption, rendered inconsistently across platforms or removed during reposting will do little to prevent manipulation.
There is also a danger that “AI-generated” becomes shorthand for “false,” damaging legitimate reporting and art, while unlabeled human-made misinformation escapes scrutiny. The more important question is not simply how a file was produced, but whether its claims can be verified and who stands behind them.
Businesses have reasonable objections. Compliance can be technically difficult when content passes through several editing tools, and smaller developers may lack the resources to implement sophisticated provenance systems. Rules that are unclear or incompatible across jurisdictions could also entrench the largest platforms, which can afford legal and engineering teams while smaller competitors struggle.
Those concerns deserve attention, but they do not justify weak rules. They argue for workable standards, support for smaller firms and enforcement that targets deliberate deception rather than harmless experimentation. Europe’s decision to provide practical guidance is a step in that direction.[2]
What accountability should mean
First, provenance information should survive ordinary sharing wherever technically possible. Second, platforms should explain labels in plain language instead of presenting obscure technical badges. Third, organizations using AI in consequential settings should retain records showing how outputs were produced, checked and corrected.
Finally, enforcement must focus on harm. A manipulated video used to defraud voters or impersonate a public official deserves more urgent scrutiny than a clearly identified fictional image. Regulators should publish decisions, measure whether labels are understood and revise the system when evidence shows that users are being misled.
Transparency is valuable only when it changes what people can know, question and challenge.
The AI Act cannot solve the information crisis alone. Newsrooms still need verification, platforms still need enforcement, and audiences still need media literacy. But the law can establish a baseline: synthetic content should not masquerade as unaltered reality, and companies should not be allowed to treat disclosure as a complete answer to foreseeable abuse.
That is the standard we should demand. Label the content—but also preserve its provenance, explain its limits and hold accountable those who use it to deceive. Anything less risks turning transparency into decoration.