Navigating the EU AI Act: A Guide for Game Developers on Deepfake Transparency

As the gaming industry integrates generative artificial intelligence (AI) at an unprecedented pace—from procedural world-building to dynamic NPC dialogue—a new regulatory frontier has emerged. Antonia Gold, a legal expert at Sheridans, has provided critical insights into the European Union’s Artificial Intelligence Act, which introduces stringent transparency obligations effective August 2, 2026. For game developers worldwide, the legislation represents more than just a compliance hurdle; it is a fundamental shift in how creative, synthetic media is categorized and disclosed to the public.

The Core Mandate: Understanding the EU AI Act

The EU AI Act is designed to mitigate the risks associated with "deepfakes"—a term that, in legal parlance, is significantly broader than the pop-culture definition of illicit celebrity videos. Under the Act, a deepfake is defined as AI-generated or AI-manipulated content that depicts existing persons, objects, places, or events in a way that falsely appears authentic or truthful.

While the definition sounds narrow, the implications for the games industry are vast. The regulation covers all media formats, including audio, video, and imagery. Crucially, the Act possesses extraterritorial reach: any studio—regardless of its headquarters—must comply if its products or promotional materials containing AI-generated content are distributed or made accessible within the European Union.

Chronology of Implementation

• August 1, 2024: Initial provisions of the EU AI Act came into force, setting the stage for systemic oversight.
• July 21, 2025: The European Commission issued specific guidance regarding the practical application of transparency rules for generative AI.
• August 2, 2026: The full scope of transparency obligations, including mandatory labelling for specific AI-manipulated content, becomes legally enforceable.

Why Games Are Not "Deepfakes" (Mostly)

For months, there was widespread anxiety within the development community that the use of generative AI for skyboxes, textures, or procedural dialogue might be prohibited or necessitate intrusive "AI-generated" watermarks on every asset. However, recent guidance from the European Commission has provided much-needed relief.

The Commission has clarified that common gaming practices—such as generating fantasy landscapes, non-human entities, or stylistic assets for fictional worlds—do not typically trigger the "deepfake" label. The underlying logic is simple: a player navigating a neon-lit cyberpunk city or a dragon-filled medieval forest understands that they are experiencing a work of fiction. Because the content does not claim to represent a "real" place or a "truthful" event, it falls outside the intended scope of the deepfake regulations.

The Grey Area: When Fiction Mimics Reality

Despite the exemptions for creative fantasy, a significant "grey area" persists. The risk of regulation increases proportionally with the degree of photorealism and the use of real-world references. Studios must exercise extreme caution when:

1. Recreating Historical Figures: Using AI to generate a photorealistic version of a deceased leader or historical personality.
2. Simulating Celebrity Personas: Implementing AI-generated voices or likenesses of real-world actors, athletes, or musicians.
3. Depicting Real Events: Using generative AI to create news-style footage or sports commentary within a game that mimics real-world broadcasts.

The "deception test" remains the primary legal benchmark: Could a reasonable observer believe they are seeing or hearing something real? If the answer is yes, the studio is likely required to provide clear disclosure.

Beyond the Game Client

A critical oversight for many studios is the assumption that the game client is the only point of concern. The regulation explicitly extends to:

• Marketing Trailers: Pre-rendered videos used for promotion.
• Out-of-Home Advertising: Billboards and digital displays.
• Influencer Campaigns: Sponsored content that uses AI to mimic reality.
• Social Media Assets: Short-form video content created via AI tools to hype game releases.

If a trailer utilizes AI to generate a photorealistic sequence of a real-world city that looks indistinguishable from archival footage, it must be labelled as AI-generated, even if the game itself is clearly stylized.

The Two-Tier Labelling System

When content is determined to be a "deepfake," the EU AI Act mandates a two-tier approach to disclosure. This is not a "one-size-fits-all" mandate, but rather a flexible framework that accounts for context:

Tier 1: Creative and Fictional Works

This tier is designed for content where the artistic intent is obvious. For most game developers, this is the default category. The requirement here is for disclosure that the content has been artificially generated or manipulated. While the Commission provides optional visual designs for these labels, the exact method of delivery (e.g., a splash screen, a watermark, or a metadata tag) remains subject to a "case-specific assessment." Studios must consider the audience, the medium, and the prominence of the label to ensure transparency.

Tier 2: Public Interest and High-Risk Content

This tier applies to content that could realistically be mistaken for genuine journalistic or public information. This is rarely the primary focus of entertainment software, but it could impact "serious" games, educational simulations, or historical documentaries produced within a game engine. These require more stringent, overt, and unmissable labelling.

Official Guidance and Compliance Strategies

The European Commission’s recent focus on the gaming sector acknowledges that the industry operates on a spectrum of reality. For studios, the best defense against regulatory overreach is a robust, internal AI governance policy.

Developing an Internal AI Policy

To ensure compliance by the 2026 deadline, studios should formalize their AI workflows:

1. Categorization: Audit all AI tools in use. Determine which assets are "creative/fictional" and which approach "photorealism/realism."
2. Training: Legal departments must work with creative leads to ensure that the "deepfake" definition is understood at the team level. An artist should know if their asset requires an AI-disclosure tag before they export it.
3. Documentation: Keep a ledger of where and how AI is used. This is invaluable in the event of an audit.
4. The "Reality Check" Questionnaire: Every time an AI asset is finalized, teams should ask:
   • Does this depict a real person or place?
   • Could an audience mistake this for authentic footage?
   • Is this being used in a marketing context where it could be misinterpreted?

Implications for the Future of Development

The EU AI Act does not signal the end of AI in gaming, but it does signal the end of the "wild west" era of AI implementation. Studios that embrace transparency as a part of their brand identity will likely face fewer hurdles than those that attempt to obfuscate their use of synthetic assets.

By establishing clear, internal guardrails today, developers can avoid the retroactive costs of redesigning marketing campaigns or re-releasing assets in the lead-up to August 2026. The key takeaway for the industry is that the law is not trying to ban the technology; it is trying to ensure the audience is empowered with the knowledge of what is real and what is synthetic. As the line between digital fantasy and physical reality continues to blur, transparency will become the ultimate marker of a studio’s integrity.