Reshaping the game - 2026 report
The EU AI Act: a compliance roadmap for the games industry
The AI Act reaches games on two tracks - what your game does to players, and which models you deploy - and its transparency duties came into force on 2 August 2026 (Article 50 AI Act, Regulation (EU) 2024/1689).
Section 4 in five theses
- AI systems that exploit player psychology to drive purchases can cross into the banned zone (Article 5 AI Act). Monetisation mechanics deserve the first audit.
- From 2 December 2026, the Act also prohibits AI creation tools without effective safeguards against sexualised imagery of real persons and child sexual abuse material (Article 5(1)(ba), (bb) AI Act as amended).
- Emotion recognition remains the industry's main high-risk trigger (Annex III AI Act), but only where a system infers emotions from biometric data. Biometric categorisation carries a notification duty even below that threshold (Article 50(3) AI Act).
- As of 2 August 2026, players must now be informed when they interact with AI systems or see synthetic content (Article 50(1), (2), (4) AI Act). The final Guidelines of 20 July 2026 name video games expressly at six points, grant targeted reliefs, and existing generative systems have a marking grace period until 2 December 2026. Marked output must also be detectable.
The provider question has two layers: a studio that builds a feature around a third-party model is usually the provider of that AI system and owes the transparency duties (Article 3(3) AI Act) - and a studio that counts as a GPAI model's "provider" additionally owes documentation, a copyright policy and a training-data summary (Article 53 AI Act).
Why the Act is a games problem
The old reflex - "pathfinding AI fits no risk category, so the Act ignores games" - fails for generative and agentic systems.
Timeline first
The adopted Omnibus Regulation delays high-risk compliance to 2027/2028 and grants existing generative systems a marking grace period until 2 December 2026; the remaining transparency duties applied from 2 August 2026.
Prohibited practices: monetisation mechanics and the new content prohibitions
The Act bans AI-driven player manipulation and the exploitation of children (Article 5(1)(a), (b) AI Act); from 2 December 2026 it also prohibits AI systems without effective safeguards against sexualised imagery of real persons- creation tools join monetisation mechanics on the audit list.
High-risk systems: emotion recognition and its boundary
Emotion recognition is the industry's main high-risk trigger, but only where a system infers emotions from biometric data - voice and face are in, gameplay telemetry as such is not.
Transparency: Article 50, the Final Guidelines and the Code of Practice
Tell players when they interact with AI or see synthetic content; the final Guidelines name video games repeatedly and grant the industry three tailored reliefs.
Wrapping a model: when the studio provides the AI system
A studio that builds a feature around a third-party model is usually the provider of that AI system - the duty is easy to acquire, and reliance on the model's marking is permitted but responsibility is not transferable.
Contacts
© 2026 Bird & Bird