AI-Generated Game Assets After the 2025 Copyright Office Report: Copyright, Platform Policies, and IP Ownership for Studios

Game studios are integrating generative AI at every stage of production — concept art from Midjourney, background music from Suno, NPC dialogue from ChatGPT, and gameplay code from Copilot. But the legal framework for owning, registering, and shipping those assets shifted dramatically in 2025. The U.S. Copyright Office released its Part 2 Copyrightability Report in January 2025, and the D.C. Circuit Court of Appeals affirmed the human-authorship requirement in Thaler v. Perlmutter two months later. Meanwhile, Steam's AI disclosure policy is now enforced, and Apple and Google Play have both published AI-specific requirements. If your studio is shipping an AI-assisted title without a clear IP strategy, you're taking on risk you may not fully understand.

We've written about AI-generated game assets and the legal risks studios face before. This article updates that guidance for the post-Part 2 Report landscape, with a focus on four areas studios ask us about most: copyright registration, platform disclosure, the IP ownership chain across AI tools, and work-for-hire complications.

1. Can You Copyright AI-Generated Game Assets? The January 2025 Framework

The single most important document for any studio using generative AI is the U.S. Copyright Office's Part 2 Copyrightability Report, released January 29, 2025. The Report addresses head-on whether outputs created using generative AI can be registered for copyright protection — and the answer is: it depends entirely on the degree of human creative control.

The Core Rule: Human Authorship Determines Copyrightability

The Copyright Office affirmed that "the outputs of generative AI can be protected by copyright only where a human author has determined sufficient expressive elements." This can include situations where a human-authored work is perceptible in an AI output, or where a human makes creative arrangements or modifications of the output. What it does not include is "the mere provision of prompts."

For game studios, this means the critical question is: who is making the creative decisions, and how much? If a concept artist writes a detailed prompt, iterates on the output, manually edits the result, composes it into a scene, and applies their own color grading, the human creative contribution is substantial. If a producer types "fantasy castle on a hill" into Midjourney and uses the first result as a loading screen, the human contribution is minimal — and the copyright claim is weak or nonexistent.

Crucially, the Report also confirms that "the use of AI to assist in the process of creation or the inclusion of AI-generated material in a larger human-generated work does not bar copyrightability." This is the key nuance: a game as a whole can still be copyrighted even if individual assets within it were AI-generated. The copyright extends to the human-authored elements — the game's code, the arrangement of assets, the level design, the narrative structure — but not to the AI-generated components standing alone.

Thaler v. Perlmutter: The D.C. Circuit Weighs In

On March 18, 2025, the D.C. Circuit Court of Appeals affirmed in Thaler v. Perlmutter that human authorship is a "bedrock requirement" for copyright registration. Dr. Stephen Thaler had sought to register a copyright listing his "Creativity Machine" AI system as the author. The court rejected this, noting that the Copyright Act's provisions on ownership, inheritance, and copyright duration all presume that authors are human beings.

However, the court acknowledged that "there might be disagreements over how much AI contribution is permissible for a work to still be considered authored by a human." Those line-drawing questions — exactly what the Copyright Office's Part 2 Report addresses — were not before the court. The practical takeaway: Thaler settles that an AI system cannot be an author, but the harder question of how much human modification suffices remains a fact-intensive, case-by-case determination.

What This Means for Studio Registration Strategy

When registering your game with the Copyright Office, you must disclose AI-generated content and specify which portions are human-authored. The Copyright Office's March 2023 registration guidance (which the Office plans to supplement) requires applicants to identify AI-generated material and exclude it from the copyright claim. For studios, this means:

• Document the human creative process for each asset: who made decisions, what modifications were applied, and how much of the final output reflects human judgment.
• Register the game as a compilation or collective work, claiming copyright in the selection, arrangement, and human-authored elements — not in individual AI-generated assets.
• For AI-assisted code, review our Copyright Office playbook for registering AI-assisted code, which walks through the disclosure requirements in detail.

2. Platform Policies: What Steam, Apple, and Google Play Require

Each major platform has now published AI-specific rules. Studios shipping AI-assisted games must comply with all applicable policies simultaneously — and they are not identical.

Steam: Mandatory AI Disclosure Survey

Valve's AI disclosure policy, published via Steamworks, requires developers to complete an AI disclosure survey when submitting a game. The policy distinguishes between two categories:

• Pre-Generated AI Content: Any content created using AI tools during development — art, code, music, text, voice acting. Developers must disclose that AI was used and confirm the content does not infringe existing copyrights or other IP.
• Live-Generated AI Content: AI that generates content dynamically during gameplay — for example, an NPC that generates real-time dialogue using an LLM. Developers must disclose this and describe the guardrails in place to prevent harmful or illegal outputs.

Valve reviews these disclosures as part of the store page review process. Games that fail to disclose AI content, or whose AI-generated content is found to infringe third-party IP, can be removed. The policy is now enforced — it is not advisory.

Apple App Store: AI Safety and Content Requirements

Apple's App Store Review Guidelines address generative AI under the performance and safety sections. Apps that incorporate generative AI must implement content filtering, age-appropriate restrictions, and clear user-facing disclosures. Apple requires that apps with AI features provide mechanisms for users to report offensive AI-generated content. The guidelines also prohibit apps from generating content that is illegal, harmful, or that infringes IP rights.

For game studios, the key Apple requirements are: (1) build content filters into any live-generative AI features, (2) provide in-app reporting mechanisms, and (3) ensure AI-generated assets do not reproduce copyrighted material. Apple's review process will reject apps that fail to demonstrate these safeguards.

Google Play: Generative AI Content Policy

In June 2024, Google Play published its generative AI content policy. Apps with generative AI features must: (1) prohibit and prevent the generation of restricted content, (2) provide users with a way to report or flag offensive AI-generated content without leaving the app, and (3) be rigorously tested to ensure outputs align with Play's policies. Google also recommends that developers document their AI testing, as Play "may ask to review it in the future."

Google's policy also covers AI-generated advertising: studios must ensure marketing materials accurately represent the app and that ads meet Google's App Promotion requirements.

3. The IP Ownership Chain: What Rights Survive When You Use AI Tools?

Even if a platform accepts your game, the deeper question is: what do you actually own? Copyright is only one layer. The other is contractual: what rights does each AI tool's terms of service grant you, and what do they withhold?

Midjourney: Commercial Use Tied to Subscription Tier

Midjourney's Terms of Service grant different commercial rights depending on subscription tier. Paid subscribers generally receive commercial usage rights to the images they generate. However, Midjourney's terms also include broad license grants back to the platform — Midjourney retains the right to use subscriber-generated content for purposes including training its models. This creates a tension: you may have commercial use rights, but you don't have exclusivity, and Midjourney's license to reuse your outputs could undermine trade-secret or confidentiality claims.

For studios, the practical risk is that concept art generated through Midjourney may not be protectable as a trade secret if the platform retains usage rights. And because the Copyright Office's Part 2 Report limits registration to human-authored expressive elements, the raw Midjourney output may not be separately copyrightable unless a human artist substantially modifies it.

Suno: Commercial Rights for Paid Users, Limitations on Free Tier

Suno's Terms of Service distinguish between free and paid tiers. Free-tier users receive only non-commercial rights to their outputs. Paid subscribers get broader commercial usage rights. However, Suno's terms require users to represent that they have all rights to any input material they provide, and Suno disclaims responsibility for whether outputs infringe third-party rights.

This is a critical gap: Suno does not warrant that its outputs are free from third-party copyright claims. If Suno's model was trained on copyrighted music (a question currently being litigated in cases we've covered for musicians), a studio that uses Suno-generated music in its game could face infringement claims from original rights holders. The studio's contractual rights from Suno do not shield it from third-party copyright liability.

ChatGPT and AI-Generated Dialogue

OpenAI's terms grant users ownership of the outputs they generate, subject to content policy restrictions. But the Copyright Office's framework means that AI-generated NPC dialogue — no matter how sophisticated — is unlikely to be independently copyrightable unless a human writer substantially shapes, edits, and arranges it. A studio that relies on raw ChatGPT output for quest dialogue may find those lines unprotected: a competitor could copy them without infringing the studio's copyright.

The ownership chain therefore has three layers: (1) contractual rights from the AI tool (what the tool's TOS allows you to do), (2) copyright registration (what the Copyright Office will protect), and (3) third-party infringement risk (whether the AI tool's training data creates exposure). Studios need to analyze all three for every AI tool in their pipeline.

4. Work-for-Hire Complications When Contractors Use AI Tools

Many studios engage freelance artists, composers, and writers. When those contractors use AI tools — often without the studio's knowledge — the IP ownership chain breaks down in two ways.

First, the work-for-hire doctrine assumes a human author is creating the work. If a contractor uses Midjourney to generate concept art and then lightly touches it up, the "work" may not be fully human-authored, which means the studio's copyright claim is weaker than it assumes. The contractor may not even realize this is an issue — but the studio bears the risk.

Second, AI tool terms of service are typically agreed to by the individual user (the contractor), not the studio. If a freelancer uses a free-tier Midjourney account (non-commercial only) or a free-tier Suno account, the studio may receive assets with no commercial rights at all — even though the studio paid for the work.

The fix is contractual. Studios should require contractors to: (1) disclose any use of generative AI tools, (2) represent that they hold the necessary commercial rights to all AI tool accounts used, (3) warrant that the final delivered work reflects sufficient human creative contribution to be copyrightable, and (4) indemnify the studio for any IP claims arising from AI-generated content. For a broader framework, our guide to IP ownership of AI-generated work product covers the contractual architecture in more detail.

5. Trade Secret vs. Copyright for AI-Generated Game Code

Code presents a unique challenge. Unlike art assets, which are visible and distributed as part of the game, source code can be kept confidential — making trade secret protection a viable alternative or supplement to copyright. The Copyright Office's Part 2 Report and the Thaler decision leave open the question of how much human modification to AI-generated code is needed for copyrightability. For now, studios should consider a dual strategy:

• Trade secret: Keep AI-assisted code confidential. If you don't distribute source code, trade secret protection applies regardless of whether a human or machine wrote the lines. This is particularly valuable for proprietary game engine modifications and server-side code.
• Copyright: Register the human-authored portions of the codebase. Document which modules were written by humans, which used AI assistance, and the degree of human review and modification. The Copyright Office requires disclosure of AI-generated content, so be specific about what is claimed and what is excluded.

Trade secret protection has a significant advantage here: it does not require human authorship. A studio can protect the confidentiality of its codebase — including AI-generated components — as long as the studio takes reasonable steps to maintain secrecy. The trade-off is that trade secret protection is lost if the code is publicly disclosed, whereas copyright protection persists.

Actionable Next Steps

For studios shipping AI-assisted games in 2026, here's what we recommend doing before launch:

1. Audit your AI pipeline. Catalog every AI tool used in production, the subscription tier, and the contractual rights granted. Identify which outputs are raw AI-generated and which have substantial human modification.
2. Strengthen human creative documentation. For assets you want to copyright, document the human creative process: iteration logs, edit histories, before/after comparisons. This evidence will be essential if the Copyright Office questions a registration or if a challenger disputes your claim.
3. Complete platform disclosures early. Fill out Steam's AI disclosure survey during the store page setup process, not after. Build content-reporting mechanisms for Apple and Google Play compliance. Test live-generative features for policy-violating outputs.
4. Update contractor agreements. Add AI disclosure and warranty clauses to all freelancer and vendor contracts. Require contractors to confirm they hold commercial-tier subscriptions for any AI tools used.
5. Consider trade secret protection for code. For server-side and proprietary engine code, evaluate whether trade secret protection is more robust than copyright for AI-assisted modules. Implement access controls and confidentiality agreements to preserve the trade secret.
6. Register strategically. File copyright registrations for the human-authored elements of your game — the compilation, the code you wrote, the narrative you designed — and disclose AI-generated content as required. Don't claim copyright in raw AI outputs.

The legal landscape for AI-generated game assets is still evolving. The Copyright Office has committed to supplementing its registration guidance and releasing Part 3 of its AI Report (addressing training data and licensing). The Supreme Court declined to review Thaler v. Perlmutter, leaving the D.C. Circuit's human-authorship ruling as the controlling appellate precedent. Studios that build a clear IP strategy now — documenting human creative contributions, complying with platform policies, and managing the contractual ownership chain — will be positioned to ship with confidence. Those that don't may discover, at launch or in litigation, that their "assets" aren't actually theirs.