The EU AI Act’s deepfake transparency rules present an interesting challenge for game studios. The obligation is broader than the popular understanding of a deepfake, extending beyond fake videos of politicians, to a wide range of AI-generated and AI-manipulated content. This has prompted concern among many game studios that AI-generated assets may be prohibited – yet recent guidance suggests that several common uses of AI in games are unlikely to fall within its scope. The real complexity lies in the grey areas between fantasy and realism.
The EU AI Act’s transparency obligations, including the requirement to label certain AI content, are here and apply from 2 August 2026. Notably, it covers all media types (audio, video and images).
As with most EU digital regulations, it applies beyond Europe, which means studios that are based outside the EU could also be caught. The labelling obligation is triggered where a studio is located outside of the EU, creates an AI deepfake and expects it to be distributed in the EU including by incorporating it in a game that is available in the EU or by posting the content online.
Welcome clarity for the games industry
The good news is that the recent European Commission guidance, released on 21 July, provides last-minute but helpful colour for game studios. For once, games are not treated as an afterthought of a digital regulation. It makes clear that common uses of AI within games are unlikely to be a deepfake. Specifically:
- AI voice replication for fictional characters when there is no deception as to the identity of the narrators
- AI-generated fictitious environments (for example, fictitious forests and castles)
- AI-generation or manipulation of background scenes, special effects, or technical pre- and post-processing.
This makes sense. The deepfake obligation is trying to capture content that creates a false impression of authenticity and truthfulness. A fantasy underwater world or medieval battle with dragons is unlikely to appear to be a real place or event.
Where the grey area begins
However, there are circumstances in which game content does raise deepfake questions. Examples include the recreation of celebrity voices, the depiction of real sports events, or the inclusion of replicas of real historical figures. A high degree of photorealism renders it more likely that such content will be considered a deepfake.
It is also worth remembering that the obligation goes beyond the game itself. Trailers, billboards, advertisements, influencer campaigns and other social media content could all trigger it.
Your legal obligation extends beyond the game itself. Trailers, billboards, advertisements, influencer campaigns and other social media content are all subject to it
The key question is whether the AI content could lead people to believe they are seeing or hearing something real. The AI Act defines a deepfake as AI-generated or AI-manipulated content that resembles existing persons, objects, places, entities or events and would falsely appear to be authentic or truthful.
Importantly, it is enough for simulated content to resemble someone or something that exists, can plausibly exist or could have plausibly existed in reality to meet that aspect of the definition. The rationale is that such content could be a source of deception.
Gaming examples
The distinction is often easier to understand in practice. The following examples illustrate the types of content that are less likely, and more likely, to trigger the deepfake labelling obligations.
Likely not a deepfake:
- AI-generated voice acting for a witch, alien or robot character
- AI-generated fantasy worlds with floating islands, glittering caves or underwater kingdoms
- AI-assisted lip-syncing or body language for fictional characters
- AI-generated fictional brands or products appearing in-game.
Could be a deepfake:
- A football game recreating a real athlete’s voice using AI
- A racing game using a realistic AI-generated version of a real driver in promotional materials
- A documentary-style game trailer depicting real-world political events through AI-generated footage
- A billboard featuring AI-generated images of real musicians.
The two-tier approach to labelling
Where an AI-deepfake has been generated, it must be disclosed that the content has been artificially generated or manipulated. In practice, in most cases, this will mean adding a visual label. The Commission has made optional label designs available to use.
There are two tiers here:
- If the deepfake forms part of an evidently artistic, creative, satirical or fictional work, the disclosure obligation must only be met in a manner that does not interfere with the display or enjoyment of the work.
- If the deepfake does not form part of an evidently artistic, creative, satirical or fictional work, it must be labelled in a clear and distinguishable manner at the latest at the time of the first interaction or exposure to the content.
The European Commission’s guidelines expressly recognise that gaming imagery can fall within the first more flexible tier for creative/ fictional work. The guidance is less specific on how such content should be labelled, but notes that relevant factors include the nature of the content, the audience and the context. This suggests that the Commission expects a case-specific assessment rather than a one-size-fits-all approach, giving studios some flexibility. The takeaway is that context matters. The same AI-generated content may be treated differently depending on whether it appears within a creative work or is used in promotional material, with the latter more likely to have stricter rules on how and when the label must be shown.
Practical steps towards compliance
A sensible first step towards compliance is a tailored AI content policy that sets out when staff can use AI to generate or manipulate content, when such content needs to be labelled and how to label such content. This should be grounded in the reality of the specific studio and reflect the nature of the games they develop and the AI they use. Ideally, the staff actually creating AI-generated content should be trained on the policy so that it becomes a practical tool for day-to-day decision-making, rather than a document that sits forgotten in a dusty corner of the intranet after the lawyers have labelled it “final”.
That policy should lead staff through a series of practical questions, with the aim of capturing the nuance of the rule and avoiding over- or under-labelling:
- Is this depicting a real or plausibly real person, object, place or event?
- Is the content highly realistic?
- Could audiences believe it is authentic?
- Will it be distributed in the EU?
The Commission’s guidance goes some way towards demystifying deepfake labelling for the games industry. Many AI-powered features found in games are unlikely to qualify as deepfakes. For most studios, the key challenge is developing the judgment needed to determine when AI-generated content tips into territory that requires disclosure.
