The Case for Mandatory AI labeling

Why transparency must become a safety standard in the synthetic age.

AI-generated content is saturating our feeds, our politics, and our sense of reality. Yet unlike every other product that touches public life, from food to finance, synthetic media carries no label of origin. If society is to preserve trust in the digital sphere, it's time to treat AI transparency as infrastructure, not ideology.

The Synthetic Moment

It took just one image of the Pope in a white designer puffer coat to expose how fragile our perception of truth has become (McElwee, 2024). Created with text prompt and a diffusion model, the image went viral, fooling millions. The episode wasn't malicious, merely illustrative: in 2025, authenticity is no longer a given, it's a guess.

We have entered what might be called the synthetic turn, where machines generate plausibility faster than humans can verify it. Texts, images, and voices now flow from models trained on oceans of data, yet arrive without any trace of provenance. The result is a new kind of uncertainty: not knowing whether to believe our eyes.

The Missing Label

Every mature technology that poses risk to the public eventually develops its own language of safety. Appliances have electrical certifications; food carries ingredient lists; financial products disclose terms. These signals don't limit innovation, they enable it, by creating trust.

AI has skipped that step. The world is now filled with algorithmically generated material that looks, sounds, and reads like the real thing, but with no standard mechanism for disclosure. Whether a marketing video, a journalistic image, or a synthetic voice note, the audience is left to guess.

"An electrical label doesn't stifle creativity; it prevents electrocution. AI content deserves the same principle."

Without a consistent tag of origin, the boundary between error, artifice, and manipulation collapses. Labeling should not be optional or decorative, it should be mandatory, the digital equivalent a safety certification.

Europe's Quiet Revolution

If a history is a guide, the first serious push for AI transparency will come from Europe. The EU AI Act (Kosinski and Scapicchio, 2024) and Digital Services Act (Turillazzi et al., 2023) together require that "synthetic or manipulated media that could mislead the public" be clearly disclosed.

Critics call the language imprecise. But as with the GDPR, the real power lies in precedent: once large platforms build compliance systems for Europe, those systems often propagate globally. Maintaining different standards for each market is inefficient. Europe doesn't just regulate, it exports governance.

By embedding provenance and disclosure into law, the EU could make transparency the default expectation for the rest of the world.

The Myth of Self-Regulation

The tech industry insists it can police itself. History says otherwise. From railroads to finance to social media or pharmaceutics (OECD, 2015)(Zetterqvist and Mulinary, 2023), self-regulation always arrives after the damage is done.

Generative-AI companies are experimenting with voluntary watermarking and provenance tools like C2PA (Coalition for Content Provenance and Authenticity). Yet without legal enforcement, such tools remain half measures, easy to remove, ignore, or fake. Transparency benefits the public, but opacity benefits profit.

Self-regulation will persist right up until the first deepfake catastrophe: the forged confession, the fabricated military order, the falsified courtroom video. Then, as always, governments will legislate in crisis mode.

The question is whether we can legislate before the disaster instead.

Labelling as Infrastructure

Labelling is not censorship. It does not restrict creation or distate aesthetics; it defines context. A credible system would have three simple components:

1. Provenance by design: AI tools embed cryptographic metadata identifying how work was generated or altered.
2. Platform Disclosure: Social networks and publishers preserve that metadata and display visible cues ("AI-generated", "AI-assisted", "verified human origin").
3. Regulatory Oversight: Governments enforce disclosure in high-impact contexts; politics, advertising, education, and journalism, where truth matters most.

"Knowing an image is synthetic doesn't make it less valuable. It makes it more interpretable.

This approach mirrors other safety regimes: open technical standards, corporate accountability, and selective enforcement. The label becomes part of the content's metadata, a quiet companion, not an intrusive banner.

Rebuilding Trust

Labelling alone will not inoculate society against misinformation, but it's a prerequisite for resilience. Today, the liar's divident, the ability to dismiss real evidence as fake, thrives in the absence of provenance (Schiff et al., 2025). When truth and falsehood are visually identical, cynicism becomes rational.

Transparency reverses that logic. It re-establishes asymmetry: honesty becomes easier than deceit. Creators gain legitimacy for disclosing their methods; audience regain confidence in what they see. Over time, provenance metadata could serve as the padlock icon of the 2030s: an ambient indicator of authenticity, invisible until needed.

Labeling also protect legitimate AI art and design from moral panic. When disclosure is standardized, creative use of algorithms can flourish without being conflated with deception.

The Label as a Civic Compact

Artificial intelligence has joined the ranks of authors, artists, and reporters, it now participates in the making of culture. But participation implies responsibility. A labeling standard is not a bureaucratic nuisance; it's a social contract.

If we can demand that food products list allergens and that software updates carry cryptographic signatures, we can demand that digital content reveal its origin. In the synthetic century, the right to know how something was made is inseparable from the right to know whether it's real.

"Labeling AI isn't about distrusting technology. It's about giving humanity a fighting chance to keep believing what it sees."

We can either wait for the deepfake that breaks democracy, or we can build the trust infrastructure now. The choice before policymakers, platforms, and citizens is not between freedom and control, it is between transparency and entropy.

The future of credibility depends on choosing the former.

References:

McElwee, J. (2025). 'Pope decries 'crisis of truth' in AI after his own deepfake image', The Independent, 24 January. Available at: https://www.independent.co.uk/news/world/europe/pope-artificial-intelligence-deepfake-davos-b2684916.html (Accessed: 26 October 2025)

Kosinski, M., Scapicchio, M. (2024). 'What is the EU AI Act?, IBM. Available at: https://www.ibm.com/think/topics/eu-ai-act (Accessed: 26 October 2025)

Turillazzi, A., Taddeo, M., Floridi, L., & Casolari, F. (2023). The digital services act: an analysis of its ethical, legal, and social implications. Law, Innovation and Technology, 15(1), 83–106. Available at: https://doi.org/10.1080/17579961.2023.2184136 (Accessed: 26 October 2025)

OECD (2015) 'Industry Self-Regulation: Role and Use in Supporting Consumer Interests'. Available at: https://one.oecd.org/document/DSTI/CP(2014)4/FINAL/en/pdf#:~:text=The%20report%20notes%20that%20industry%20self-regulation%20%28ISR%29%20can,taken%20to%20help%20ensure%20that%20such%20initiatives%20succeed. (Accessed: 26 October 2025)

Zetterqvist, A., & Mulinari, S. (2013). Misleading advertising for antidepressants in Sweden: a failure of pharmaceutical industry self-regulation. PLoS ONE, 8(5), Article e62609. https://doi.org/10.1371/journal.pone.0062609

Schiff, K., J., Schiff, D., S., Bueno, N., S. (2025). The Liar’s Dividend: Can Politicians Claim Misinformation to Evade Accountability?, American Political Science Review, February, 119(1):71-90. Available at: doi:10.1017/S0003055423001454 (Accessed: 26 October 2025)