Reshaping the game - 2026 report
Final recommendations: navigating a new legal landscape
Tier the IP, think rights and transparency together, audit against the AI Act, moderate before the first incident, and treat transparency as a brand decision, tailored to each company's games, players and partners
Apply a tiered IP strategy based on content sensitivity
Not all game content carries the same legal and strategic weight (from an isolated IP perspective). The more central and distinctive a creative asset is to the game's identity, core characters, narrative arcs, visual style, the greater the need for demonstrable, documented human authorship, if the company wants to own and enforce it.
By contrast, content that is repetitive, peripheral or systemically generated, such as item descriptions, background assets and filler dialogue, can be produced or enhanced with generative AI at lower risk, provided appropriate safeguards and review mechanisms are in place. Where exactly each company draws its lines is a business decision with several defensible answers (Section 3.6) and what no path spares is the documentation that secures whichever assets the company chooses to keep enforceable.
Think IP and transparency as one strategy
New in this edition, because the two threads have converged: the decision of how an asset is made is simultaneously a rights decision and a transparency decision. It determines whether copyright arises and can be licensed (Section 3.2), whether storefront disclosures and Article 50 duties trigger (Sections 4.5, 5.1, 6.4), and what players, publishers and platform gatekeepers will see and expect (Sections 2.3, 6.3).
Companies that decide these questions separately, the pipeline here, the store form there, the press statement somewhere else, create exactly the inconsistencies that courts, platforms and communities punish. Companies that decide them together answer four exposures with one decision. The weighting differs by company: a franchise builder weighs protectability heavily, a live-service studio weighs disclosure architecture, an indie developer weighs community expectations – there is no default.
Conduct proactive audits for AI Act compliance
Not all game content carries the same legal and strategic weight (from an isolated IP perspective). The more central and distinctive a creative asset is to the game's identity, core characters, narrative arcs, visual style, the greater the need for demonstrable, documented human authorship, if the company wants to own and enforce it.
By contrast, content that is repetitive, peripheral or systemically generated, such as item descriptions, background assets and filler dialogue, can be produced or enhanced with generative AI at lower risk, provided appropriate safeguards and review mechanisms are in place. Where exactly each company draws its lines is a business decision with several defensible answers (Section 3.6) and what no path spares is the documentation that secures whichever assets the company chooses to keep enforceable.
Invest in content moderation for UGC before the first incident
New in this edition, because the two threads have converged: the decision of how an asset is made is simultaneously a rights decision and a transparency decision. It determines whether copyright arises and can be licensed (Section 3.2), whether storefront disclosures and Article 50 duties trigger (Sections 4.5, 5.1, 6.4), and what players, publishers and platform gatekeepers will see and expect (Sections 2.3, 6.3).
Companies that decide these questions separately, the pipeline here, the store form there, the press statement somewhere else, create exactly the inconsistencies that courts, platforms and communities punish. Companies that decide them together answer four exposures with one decision. The weighting differs by company: a franchise builder weighs protectability heavily, a live-service studio weighs disclosure architecture, an indie developer weighs community expectations – there is no default.
Champion transparency as a core brand value
Beyond strict legal requirements, being transparent with players about the use of AI is central to building and maintaining trust. This includes clear communication in the terms about how AI impacts game mechanics, explicit information on how player data is used to train or personalise AI systems, and in-game notices when players interact with AI-driven characters.
The empirical picture (Section 2.3) adds a hard edge to this soft factor. Inconsistency between a company's statements and what players discover is the single most reliable backlash trigger. Whatever AI posture a company chooses, extensive, selective or none, the posture, the disclosures and the pipeline should tell the same story.
By balancing the potential of generative and agentic AI with legal diligence and a clear-eyed view of player expectations, the games industry can navigate this new frontier so that its future is not only innovative but also safe, fair and respectful of player rights. Which path is right, how much AI, on which assets, with which disclosures, differs for every company; the frameworks in this guide are the map, not the route.
Annex - Where the AI Act's secondary legislation speaks about games
The Commission's guidance mentions games rarely but decisively; this annex collects every explicit reference and what each one settles.
The AI Act itself never mentions video games. Its secondary layer does, sparingly, and at points that decide questions the Act leaves open. The register below lists the explicit references as of 25 July 2026. Paragraph numbers for the Article 50 Guidelines follow the final version of 20 July 2026 (C(2026) 5054 final); the earlier draft numbered these passages differently.
Two documents contain no games reference at all: the Code of Practice on Transparency of AI-Generated Content and the GPAI copyright code. Their silence is not neutrality - it means the general rules apply to games without a sector-specific adjustment, which is why the marking mechanics in Section 4.5 have to be mapped onto game assets by the studio rather than read off the Code.
Guidelines on prohibited practices (C(2025) 884 final)
- Para. 66: A game using AI-enabled neurotechnology and machine-brain interfaces that trains the player's brain surreptitiously to reveal intrusive information falls under the subliminal-techniques prohibition, Article 5(1)(a) (→ Section 4.3).
- Para. 105: A game that profiles children's behaviour and drives excessive play through addictive reinforcement schedules and dopamine-like loops falls under the vulnerability prohibition, Article 5(1)(b) — with the express limit that toys, games and learning applications in general are not caught (→ Section 4.3).
- Para. 121: AI systems supporting children in learning and in games are listed as out of scope, confirming that threshold (→ Section 4.3).
Draft Guidelines on high-risk classification
- Annex III, example following para. 177: A gaming-industry system tracking body posture, facial expressions, eye closures and gaze to read player excitement, anger, frustration or amusement qualifies as emotion recognition under Annex III point 1(c) (→ Section 4.4).
- Annex I, paras. 56–60: AI as a safety component of, or as, a product under the Toys Safety Regulation (EU) 2025/2509 concerns AI-enabled merchandise, not games (→ Section 4.4).
Guidelines on Article 50 (C(2026) 5054 final)
- Example box following para. 45: Interactions with AI-enabled NPCs in a single-player video game, where the game's nature makes clear no other person can participate, fall under the obviousness exception, Article 50(1) (→ Section 4.5).
- Para. 68: Outputs used only in closed production loops — expressly including games production — benefit from pipeline relief under Article 50(2); only the final shipped output must be marked (→ Sections 4.5, 4.6).
- Para. 88: Ephemeral real-time generation consumed immediately, e.g. in video games, may be exempt from marking where marking is not technically feasible and players are made aware — live-generation relief, Article 50(2) (→ Sections 4.5, 6.4).
- Para. 107 and following example box: The place of deployment, e.g. a virtual gaming platform — a pop-up before a computer game launches telling the player their face is recorded — illustrates the notice duty, Article 50(3) (→ Section 4.4).
- Example box following para. 116: AI-generated fictitious environments such as forests and castles in video games, and AI voice replication for evidently fictional characters, are not deepfakes — Article 50(4) scope (→ Section 4.5).
- Para. 122: Audience expectations, including naming the gaming environment, count as a factor for the evidently creative or fictional character of a work — attenuated disclosure, Article 50(4) (→ Section 4.5).
- Example box following para. 124: AI-generated gaming imagery involving deepfake simulations of real, existing persons is listed as an analogous creative work — attenuated disclosure applies, but the duty remains (→ Section 4.5).
GPAI scope Guidelines
- Example to para. 17: A model trained specifically to play chess or video games is not a general-purpose AI model — GPAI classification (→ Section 4.7).
Template for the public summary of training content
- Video modality / online sources: Video games are listed among video content types, and gaming platforms among online sources — training-data summary, Article 53(1)(d) (→ Section 4.7).”
Two documents contain no games reference at all: the Code of Practice on Transparency of AI-Generated Content and the GPAI copyright code. Their silence is not neutrality - it means the general rules apply to games without a sector-specific adjustment, which is why the marking mechanics in Section 4.5 have to be mapped onto game assets by the studio rather than read off the Code.
© 2026 Bird & Bird