The EU AI Act could give creative AI a trust upgrade

This article is part of our Opinions section, where we invite industry professionals to share their views on the most pressing technology questions of our time. Here, Jie Yang, Co-Founder and CTO of Utopai Studios, reflects on the impact of the EU AI Act and Article 50. It is not a representation of compliance by Utopai Studios or any specific product.

It has been six years since the European Commission published its White Paper on AI, helping to set in motion the process that ultimately produced the EU AI Act. On Sunday, 2 August, a significant set of provisions became applicable, moving the Act from legislative framework to practical application and introducing transparency requirements with implications far beyond Europe’s shores.

The AI Act takes a risk-based approach. Its rules depend largely on what an AI system is designed to do, the risks created by that use and whether a business is developing the system or deploying it. It also includes separate obligations for providers of general-purpose AI models.

At the top of the framework are prohibited practices considered incompatible with fundamental rights. Below those are high-risk systems used in areas such as credit scoring, recruitment, law enforcement, asylum decisions and influencing elections or voting behaviour. These systems are permitted but subject to extensive compliance requirements, which will apply from 2 December 2027. High-risk AI incorporated into regulated products will follow in August 2028.

For many businesses in the creative economy, the most immediately relevant provisions are the transparency requirements contained in Article 50. They are more targeted than the requirements for high-risk systems, but they are no less important. Most became applicable this week, so businesses need to understand both whether they are covered and what they are actually required to disclose.

Summary of EU AI Act:

Who needs to act

Article 50 does not apply only to companies established in the EU. Providers and professional deployers outside the bloc can also fall within its scope when their systems or outputs are used in the EU.

The European Commission’s guidance indicates that companies outside Europe can also fall within scope where there is a relevant connection to the EU market or where AI-system outputs are used in the EU. As a result, many film companies, advertising agencies, publishers and professional creators based in the UK, US and elsewhere may need to assess their obligations when distributing content to European audiences.

The rules distinguish between two principal groups: providers and deployers.

Providers are organisations that develop an AI system, or have one developed, and place it on the EU market or put it into service under their own name. For systems that generate or manipulate synthetic audio, images, video or text, providers generally must ensure that outputs are marked in a machine-readable format and can be detected as artificially generated or manipulated.

That obligation is subject to technical feasibility, the state of the art and limited exceptions for standard editing or changes that do not substantially alter the original input or its meaning. Systems placed on the market or put into service before 2 August 2026 also have a limited transition period, until 2 December 2026, to comply with the machine-readable marking requirement.

This system-level marking is designed to support detection and disclosure as content moves through production and distribution. It is not necessarily a label that audiences will see.

Deployers, meanwhile, are organisations or professionals using AI systems under their authority. For organisations such as Utopai Studios, which operate across technology development and professional content production, it is important to understand both sides of that distinction.

What creators must disclose

Deployers are not required to place a visible label on every piece of AI-assisted content.

For image, audio and video, the principal disclosure obligation applies when the content constitutes a deepfake. The Act defines this as AI-generated or manipulated content that resembles existing people, objects, places, entities or events and could falsely appear to a person to be authentic or truthful.

The Commission’s guidance interprets “existing” broadly enough to include something that exists, could plausibly exist or could plausibly have existed. This means a photorealistic portrait of an invented person can potentially qualify, even though the individual does not exist.

Context and audience expectations matter. A documentary about a political event that includes realistic, AI-generated vox pops from people who never existed illustrates the kind of case the Commissioner’s guidance is aimed at: content that could lead viewers to believe that they are seeing real members of the public expressing genuine views is squarely the risk Article 50 targets.

A fictional film featuring aliens laughing at humanity’s inability to take a clear photograph of a UFO sits at the other end of the spectrum: content that is evidently unrealistic, where the audience has no expectation of watching an authentic event, is the kind of case the deepfake definition is generally understood not to reach.

The assessment is nevertheless case-specific. It considers the content’s realism and message, where and how it is presented, and the composition and expectations of the audience reasonably expected to encounter it.

Article 50 also covers certain AI-generated or manipulated text, but much more narrowly than is sometimes assumed. Disclosure is required when the text is published to inform the public about a matter of public interest and has not undergone substantive human review or editorial control by a person or organisation assuming responsibility for its publication.

This means a professionally edited article does not automatically require an AI label simply because an AI tool contributed to an early draft. To qualify for the exception, however, the human review must be substantive. A cursory spellcheck or grammatical review is not enough.

Where ordinary editing ends

One practical challenge will be distinguishing standard AI-assisted editing from changes that materially alter the content.

Minor grammatical corrections, spellchecking, technical compression, noise reduction, cropping, colour correction and other routine production adjustments can fall within the standard-editing exception when they do not substantially alter the input data or its meaning.

The picture changes when AI is used to replace a face, synthesise a recognisable voice, create a realistic performance that never happened, or add or remove people or objects in a way that changes the substance of an image or scene. These interventions may trigger machine-readable marking by the provider and, when the result constitutes a deepfake, audience-facing disclosure by the deployer.

The important distinction is therefore not simply between “AI-assisted” and “fully AI-generated” work. Partially manipulated content can require disclosure, while completely generated content may fall outside the deepfake definition if it is obviously fantastical and incapable of misleading its foreseeable audience.

Creators will need to make those assessments carefully and document the reasoning behind them. But they should not treat disclosure as an admission that the work is somehow less creative or less their own.

Film has incorporated new technologies for decades. Audiences may not recognise every visual-effects shot or digital intervention, but the use of those tools has never diminished the contribution of the filmmakers behind them. The craft lies in the creative choices: what story is told, what the audience sees and feels, and how every available tool is used to achieve that result.

The same principle should apply to AI.

Transparency without interrupting the story

There is a legitimate concern that prominent AI labels could pull viewers out of an experience and encourage them to focus on how something was made instead of the story being told.

The Act recognises that tension. When deepfake content forms part of an evidently artistic, creative, satirical, fictional or analogous work, disclosure is still required, but it can be presented in an appropriate manner that does not hamper the display or enjoyment of the work.

That does not automatically mean an end-credit disclosure will always be sufficient. The Commission’s guidance on Article 50 states that information should be clear and distinguishable no later than the audience’s first exposure. Depending on the work and its distribution, that could be addressed through information presented before playback, an opening disclosure, a platform interface or another appropriately designed notice that does not remain on screen throughout the programme.

The precise approach will be case-specific, but the principle is sensible: viewers should be informed without the disclosure overwhelming the work itself.

I believe this shift could do for AI what disclosure rules have done for influencer marketing. A decade ago, a simple “#ad” tag could feel like an admission of inauthenticity. Today, it is routine. The disclosure allows audiences to understand the context and then judge the content on its merits.

The same normalisation can happen with AI. As disclosure becomes more consistent, audiences can move beyond the instinct to hunt for undisclosed AI and instead evaluate whether the technology has been used responsibly, creatively and effectively.

Trust as a foundation for adoption

AI regulation does not have to come at the expense of innovation. The strongest technologies are not adopted simply because they are capable; they are adopted because people trust the companies creating them and understand the circumstances in which they are being used.

Article 50 will not resolve every concern surrounding generative AI, and it will certainly not be the industry’s final regulatory milestone. But transparency provides a foundation on which broader acceptance can be built.

The financial consequences of getting it wrong are significant. Violations can attract fines of up to €15 million or 3 percent of worldwide annual turnover, whichever is higher for undertakings, while for SMEs the lower applicable maximum applies. That should accelerate the development of clearer compliance practices throughout the creative supply chain.

But the more important consequence may be cultural. Greater transparency can help establish AI as a legitimate production tool – one that enhances human creativity rather than concealing or replacing it.

Because entertainment, advertising and digital media operate globally, the standards established by the EU will influence practices well beyond its borders. UK and US companies serving European audiences will need to adapt, and many will find it simpler to apply consistent standards across their international operations.

That is how the EU AI Act’s impact will travel: not only through regulation, but through the commercial and creative norms that develop around it.

For the creative industries, that should be seen less as a constraint than as a trust upgrade – and trust will ultimately determine how widely and successfully these technologies are adopted.