Roblox has always been a trend-driven platform, and the cycles keep getting shorter. When a format catches on, the leading experiences in the genre of that format pull enormous concurrency for a few weeks, and dozens of similar games launch while the trend is still climbing. Grow a Garden and Steal a Brainrot are probably the most prominent examples, but even recent success stories like Steal an Egg have spawned a legion of copycats. Developers who reach the market early capture most of the value, especially if they have an original spin to add to the formula. Developers who arrive late compete for what remains.
This pressure has led to much of the interest in using AI tools that can “one-shot” a game by generating the Luau scripts, UI, and a basic map from a single prompt. Used well, these tools speed up prototyping and lower the barrier to entry for small teams. Used as a substitute for substantially any “traditional” game development, these tools can create legal problems that most developers do not notice until a dispute forces the issue.
In our practice, we have seen several problems with this practice. A game built primarily from AI output may carry little or no copyright protection, which affects a developer’s ability to remove clones and to respond to infringement claims from others. Building with AI can also expose confidential information a studio would otherwise protect as a trade secret. In this post, we will walk through each issue and close with practical considerations for studios and solo developers who want to keep using AI tools without giving up the value of what they build.
Who owns a one-shot game?
Copyright protection exists in “original works of authorship fixed in any tangible medium of expression.” The word “authorship” implies that there is an “author” who has that ownership right. The courts and the U.S. Copyright Office interpret that to mean that a recognized legal entity, meaning a human or a legal entity operated by humans, must participate in the creation of the work for the work to have an “author” and receive copyright protection.
See, for example, Thaler v. Perlmutter in 2025. The Copyright Office’s January 2025 report on copyrightability directly addressed prompting: in the Copyright Office’s view, prompts alone generally do not give a user enough control over the expressive output to make that user the author.
Apply that to the one-shot workflow. If a developer’s creative contribution was purely limited to a prompt, even a long and clever one, the code, art, and level layout the tool returned may sit in the public domain. The developer can still publish the game but may not hold an enforceable copyright in most of the game.
However, this is not the whole story. Even if portions of a work is not copyrightable due to the use of prompting, copyright can still attach to the human parts of the work:
- code, art, audio, and writing a developer creates outside of the AI generated materials, particularly as represented in the game or in underlying documentation like Game Design Documents;
- meaningful human modifications to AI output, as opposed to cosmetic tweaks; and
- the developer’s creative selection, coordination, and arrangement of elements, even when some of those elements started as AI output.
Less human involvement means thinner protection, and a one-shot game sits at the thinnest end of that range. Layer on the fact that game mechanics and genre conventions cannot be copyrighted, and a developer who one-shots a trending format may own very little of their game beyond the name on the thumbnail.
Furthermore, a developer who registers the game with the Copyright Office must disclose and disclaim AI-generated material that is more than de minimis. An opposing party can challenge a registration that overclaims AI content, and that challenge tends to arrive at exactly the moment the registration matters most.
The DMCA as a sword: harder to swing than it looks
On Roblox, the DMCA takedown is the first tool most developers reach for when a clone shows up. In a trend-driven market, it is often the only realistic one: the trend will be over long before a lawsuit is. The process is fast and cheap. However, it assumes something the one-shot developer may not have: a copyright in the material the copycat took.
A valid takedown notice requires the sender to own the copyright (or act on behalf of the owner) and to state a good faith belief that the use by the alleged copycat is unauthorized. If the copycat lifted AI-generated scripts, maps, or assets that no human authored, there may be nothing to enforce. Sending the notice anyway is not a free swing. Section 512(f) of the DMCA creates liability for knowingly material misrepresentations in takedown notices.
The bigger problem shows up after a counter-notice by the copycat. Most avid AI users in the video game space know that there are certain telltale signs of heavy AI use. If a copycat who receives a DMCA takedown identifies that the original work relies heavily on AI, they may submit a counter-notice alleging that no copyright exists that they could have infringed upon.
Roblox has no obligations under the DMCA to adjudicate a dispute. If Roblox receives a valid counter notice from a copycat, Roblox will typically restore the content within a week or two unless the claimant files suit and comes back with a court order. Having to sue to defend a “one shot” Roblox game largely defeats the fast and low-cost purpose of the “one shot” structure. An expensive and lengthy litigation is exactly the opposite of that.
There is also a quieter version of the problem. AI tools do not guarantee unique output. Ten developers who prompt for “a pet simulator with eggs, trading, and rebirths” during the same trend cycle may get remarkably similar results. Evidence of similarity that would suggest copying between two human-built games may just reflect two people using the same tool in one shot games. This undercuts the inference of copying that most takedowns rely on.
The DMCA as a shield: defending what no one can explain
Now flip it. A one-shot game starts pulling numbers, and another developer files a takedown claiming infringement. With a human-built game, the defense starts with establishing the origin of the game: who wrote the code, where the assets came from, what the version history shows, etc. Independent creation is a complete defense to infringement, and a clean development record is how a developer proves it.
A one-shot game has almost no development record. The developer can show the prompt, but not where the output came from. If the tool reproduced recognizable portions of someone else’s scripts, a familiar UI layout, or an asset that tracks a third party’s model too closely, the developer may not find out until they receive a DMCA notice. By that point, the Developer will have a hard time rebutting the DMCA notice. The reality here is that AI tools, particularly those that have trained on broad large language models rather than an in-house style, are liable to produce infringing output that can be difficult to identify.
That creates a real bind for a developer trying to put together a DMCA counter-notice. A counter-notice requires the sender to state, under penalty of perjury, a good faith belief that the platform removed the content by mistake, and to consent to jurisdiction for a lawsuit. Signing that statement about code no one on the team has read line by line is very risky.
The exposure extends past the DMCA. Roblox’s terms require creators to hold the rights to what they upload. Publishers, investors, and acquirers will ask developers to represent that they own the game and that it does not infringe anyone else’s rights. Developers preparing for these transactions should also consider how IP ownership and development records affect the sale or investment process. A one-shot workflow makes those representations harder to give and harder to support in diligence.
Feeding the model: training and trade secret dilution
One-shotting rarely stops at one shot. Developers iterate: they paste in existing scripts, upload the economy spreadsheet, walk the tool through the monetization funnel, and ask it to fix bugs or rebalance progression. Every one of those inputs goes somewhere.
Many AI tools, particularly the free access versions, reserve the right to use inputs and outputs to train or improve their models, especially on free and consumer tiers. Some offer an opt-out, some do not, and providers change the defaults. Platforms with built-in AI features may also have their own data sharing settings. If a developer’s inputs train someone else’s model, pieces of that game logic can resurface in another developer’s output, effectively helping build a competitor’s next one-shot. Studios incorporating these tools into their workflows may also benefit from establishing AI governance policies that address confidentiality, approved tools, and third-party risks.
The trade secret angle is where this gets expensive. Under the Defend Trade Secrets Act and most state laws, information qualifies as a trade secret only if its owner takes reasonable measures to keep it secret. The server-side logic behind drop rates, anti-exploit systems, matchmaking tuning, and live-ops data are classic candidates. Pasting that material into an AI tool whose terms permit training works against the requirement to employ “reasonable measures” to keep information secret and opposing counsel will likely make that argument the first time the studio tries to enforce its trade secret rights.
Two related risks deserve a flag. First, if a contractor, collaborator, or partner shared code or data under an NDA, running it through an AI tool may breach that agreement regardless of what happens to the studio’s own rights. Second, if team members or contractors one-shot portions of the game on their own accounts, the studio may end up with neither the copyright (no human author means nothing to assign) nor any control over what went into the tool. Well-drafted contractor agreements address both.
Practical considerations
None of these means developers should stop using AI tools. It means using them like a studio that expects its game to be worth something. A few habits go a long way in protecting against the issues discussed:
- Using AI to draft, not to finish. Treating output as a starting point that a human meaningfully rewrites, restructures, and extends puts more human authorship in the final build and strengthens the studio’s position.
- Keeping a development record. Version control, saved prompts, and notes on which elements started as AI output (and what changed) support registration, DMCA disputes, and diligence alike.
- Reading the tool’s terms and checking settings. Knowing whether inputs train the model, opting out where possible, and favoring business or enterprise tiers with stronger confidentiality commitments all reduce exposure.
- Keeping the crown jewels out of the prompt box. Server-side logic, economy tuning, anti-cheat, and player data do not belong in a tool that can train on them.
- Registering strategically and accurately. A registration that covers the parts of the game reflecting real human authorship, with AI-generated material properly disclosed, holds up when it matters.
- Updating agreements. Contractor, collaborator, and revenue share agreements should address AI use, disclosure, confidentiality, and ownership of the work product.
- Pausing before sending or countering a DMCA notice. When AI generated the content at issue, a conversation with counsel before signing anything under penalty of perjury is worth the time.
The bottom line
One-shotting is a great way to test whether a trend has legs. It is a poor way to build an asset. The days saved up front can cost a studio the ability to stop clones, defend its own work, close a publishing deal, or keep its best systems confidential. And on a platform where trends cycle this quickly, second shots are rare.
Courts and the Copyright Office will keep refining the rules over AI use and copyright the next few years. In the meantime, the developers who come out ahead will treat AI as a collaborator rather than an author and will document the difference.
Developers building on Roblox with AI tools who want to pressure-test a workflow, contractor agreements, or IP strategy can reach out to Odin Law and Media.
View all posts by this author