Media, Entertainment & Sport
Creating real-time entertainment content
Agentic AI is reshaping the landscape of digital entertainment development, moving beyond traditional generative tools that simply create static assets or predefined content. In the next evolution - agentic entertainment development - AI systems autonomously design, run, manage and dynamically evolve parts of, and even whole pieces of content.
An agentic AI assistant could:
- act as a DJ and generate music for an event in real time;
- create personalised TV series to watch; and
- produce, host, and manage video game content.
Video games
The spectrum of entertainment content is broad, ranging from music, text, images, and videos to interactive media such as games. This analysis focuses primarily on the video game industry.
In gaming, there are ongoing concerns about delegating the creation of a game’s IP - its core asset - to AI. Some developers may experiment with agentic AI to generate and run entire casual games (e.g. board or card games), including features like player matching. However, many game providers are likely to continue building on their "frontend IP", such as storylines, worlds, or iconic characters, while using agentic AI more selectively for “backend IP” such as code.
This may involve:
- managing monetisation processes, such as in-game item sales or advertising;
- adapting gameplay in real time to player behaviour and skill level; or
- generating peripheral but content-rich elements at a scale that would be difficult or inefficient to produce manually. For example, large numbers of NPCs with real-time dialogue, dynamic environments, or responsive side quests within pre-designed narrative frameworks.
Copyright: Real-time content & IP landmines
As games evolve into dynamic worlds that generate content in real time, studios face a dual challenge: protecting AI-generated material and ensuring autonomous agents don’t infringe third-party rights as new content is created.
How can real-time AI-generated content be protected?
Fully autonomously, machine-generated story arcs, dialogue, or source code don’t often meet the threshold for copyright protection under CJEU case law (see the “Autonomous Software Development” section). As a result, it may be advisable for agentic AI pipelines to rely on a pre-cleared “design corpus” of assets, such as iconic characters, core plot elements, and key items, whose existing copyright can extend to AI-generated outputs when these elements are reproduced identically or recognisably.
This leads to a key principle: the more important the asset, the more tightly the AI agent must be constrained - potentially even to deterministic behaviour (i.e. generating a specific, predefined output). For example, the main character may need to be reproduced exactly as designed by the agentic AI, while the AI can be given more autonomy in shaping the game environment or adjusting the difficulty level.
What to consider next:
- This approach requires a clear agentic AI strategy to ensure that the game’s core assets remain protectable.
Can real-time content infringe third-party rights?
Beyond securing rights in new content, developers must also address the risks of infringement that come with agentic AI expanding virtual worlds in real time. The danger lies in the AI potentially incorporating third-party, copyright-protected material - either identically or recognisably - while generating content on the fly. For instance, an AI dungeon master might spontaneously generate a villain that looks and sounds exactly like a famous copyrighted character from another game/movie.
In such cases, content teams (or even monitoring AIs) can only intervene after the infringing material has been created and potentially shared. Because the developer has actively chosen to deploy the agentic AI, courts may attribute its actions to the game provider, depending on the specifics of the case.
As a result, the developer may be potentially held directly liable for copyright infringement, if the game provider lacks the necessary rights and no copyright exceptions apply e.g. pastiche or parody. In practice, this calls for clear constraints on how the AI behaves: define its content boundaries and ensure a minimum level of human oversight. For example, requiring human approval before releasing key assets such as new characters or plot twists. This, in turn, restricts the possibilities of real-time content.
There may also be increased focus on liability arrangements between the agentic AI providers and the game providers. A shift toward an intermediary liability model - where liability arises only if the game operator fails to act after receiving a sufficiently specific notice - may be appropriate in scenarios where players have the option to influence the outputs of agentic AI. For example, if an agentic AI allows players to generate content and they bypass safeguards to create a copyrighted asset such as a protected character from another game, principles of intermediary liability could apply. Once the games provider is notified about this manipulation, they’d need to take the infringing content down.
What to consider next:
This development requires a robust strategy in the future for all parties involved.
- Games providers: Establish copyright guardrails for real-time content generation. Identify which assets require human review before release and define contractual responsibilities between game provider and AI vendor.
- Rightsholders: Develop your own litigation strategy in cases where agentic AI uses your content in a virtual world without your consent.
Media regulations: protecting the youth
Real-time content generation by agentic AI raises significant challenges under national and EU youth protection laws. Media and child protection regimes typically divide content into three regulatory categories:
- Content that is strictly prohibited: Across the EU, certain categories of content are prohibited - for instance CSAM or terrorist content under harmonised EU instruments - while a second tier of nationally defined prohibitions, such as Holocaust denial in France or the display of unconstitutional symbols in Germany and Austria, creates a patchwork of jurisdiction-specific bans that AI-generated real-time content can violate at scale, instantaneously, and without the friction that previously limited human-driven dissemination. Under German youth protection law, this could result in fines of up to €500,000 per violation. To mitigate this risk, developers must implement content filters tailored to each jurisdiction, ensuring that the agentic AI cannot generate blacklisted material in real time.
- Age gating: Content permitted for adults but restricted for minors requires robust age-verification systems. In Europe, age classifications follow schemes such as USK (6, 12, 16, 18) or PEGI (3, 7, 12, 16, 18). Depending on the jurisdiction, game providers must ensure that agentic AI outputs align with the game’s assigned age rating. There are two main options:
- Apply the strictest applicable rating globally and implement global safeguards to prevent the AI from generating inappropriate real-time content (e.g. graphic violence, sexual themes); or
- Implement jurisdiction or zone-specific age gates, though this approach is more complex and harder to enforce consistently.
In either case, the agentic AI must be technically constrained to stay within the game’s age rating. Violations could result in regulatory action, including takedown orders and significant fines.
- Unpredictable real-time content: AI-generated content is interactive and dynamic which poses two challenges:
- In countries like Germany where interaction risks are factored into age ratings, AI-generated real-time content may require additional descriptors and a higher rating.
- Ratings typically assume that the authority (e.g. USK) has reviewed all relevant game content pre-release. However, this would not be the case if the agentic AI generates essential elements relevant for review in real time after the launch.
Unless age rating systems adapt, games using agentic AI may lack a valid or complete rating since not all content can be reviewed in advance. To avoid this, developers may need to constrain agentic AI within a defined framework rather than allowing open-ended creative output that cannot be reliably rated. This could be by relying on pre-approved narrative modules or controlled graphics and vocabulary that can be fully assessed by rating bodies.