Anthropic Blocks AI Game Clones! AI Game Development Changes with Just Two Sentences: Mixed Reactions to Anthropic's Clone Countermeasures

Claude Restricts "Make That Game" — Where is the Line Between Creation and Copying in the AI Era?

"Make the same game as that one"

Simply asking AI to do so can assemble familiar game mechanics and screens. While this is a testament to technological advancement, it is not entirely a cause for celebration for those who created the original works.

AI development company Anthropic is attempting to draw a new line in this boundary.

According to a report by The Verge on November 3, 2026, Japan time, the company updated the prompts for Claude Code to adjust them so as not to proceed in the direction of recreating specific existing games. On the other hand, it will continue to support the creation of original games.

The core of the change is just two sentences. However, behind it lies a significant issue concerning the protection of creators, the preservation of game culture, and the extent to which AI services should be entrusted with judgment.

What Has Changed — It Does Not Prohibit Game Creation Itself

The movement that drew attention was triggered by a post from X user chasm.

The user disclosed that new instructions were added to Claude Code to limit the recreation of existing games. The rules shown in the post allow for the creation of original works, games of specific genres, and those using the user's own designs and materials, while avoiding the recreation of specific existing games. It is said that even changing the name or claiming it to be original would still be subject to this rule.

However, Anthropic has not officially confirmed all the wording in the posted content.

Addy Osmani from the company explained that two sentences were added to the prompt of Claude Code running on Opus 5.5 to direct it not to recreate specific games of others. He stated that using Claude to infringe on intellectual property violates the company's usage policy.

At the same time, he explained that it is assumed that original works, genre-based games, and games using the user's own designs and materials can continue to be created.

Therefore, interpreting this change as "Claude can no longer create games" is not accurate. The aim is to continue supporting creation while curbing the direct recreation of others' works.

Can "Same Genre" and "Recreation of Specific Works" Be Easily Distinguished?

Reading only the explanation of the policy, the demarcation seems clear.

However, in actual game development, there is a wide area between referencing and recreating.

For example, when creating a game where you jump over obstacles while moving sideways, developers might think of existing works during the process of considering jump height, platform placement, and collision detection with enemies.

From there, do they construct a unique world and rules, or do they get closer to the stages, appearance, and presentation of a specific work? The intention and the end product differ between the two.

The problem is whether AI can sufficiently understand this difference through short interactions.

Whether the reason for mentioning a work's name is to seek a copy of the finished product or to explain the feel of the operation. Whether what appears to be existing material belongs to others or is something the requester owns the rights to. These circumstances may not be discernible from the prompt alone.

Anthropic itself acknowledges room for adjustment. Osmani stated that they will continue to improve to achieve the right balance and called for users to inform them if Claude rejects their own work.

This call indicates that the purpose of the policy and the actual judgment may not always align.

SNS Reaction ① — "What About Consistency with Learning Data?"

Reactions on social media to this explanation questioned not only the content of the restrictions but also the trust in AI companies.

Replies to Osmani questioned whether the handling of learning data by AI companies themselves and the consideration for intellectual property demanded of users are consistent.

This is an evaluation from the poster's side and does not certify that individual learning activities were illegal in this article. Nonetheless, there is a sense of dissatisfaction that if companies advocate for the protection of creators, they should also explain their own stance.

Users are not only looking at whether the prohibitions are reasonable but also questioning whether they can trust the explanations of the companies setting those rules.

SNS Reaction ② — Concerns About Whether Mods and Game Preservation Will Also Be Stopped

Another point of contention is the anxiety that work involving existing games will be collectively restricted.

There were voices questioning the impact on mod creation and advocating for the need to preserve old games that no longer run on modern devices. There were also concerns about whether works that allow reuse under their terms or activities permitted by rights holders would be rejected.

These indicate that the purpose of handling existing works is not singular.

Creating a similar competing work, running one's own work in a different environment, and making authorized modifications have different purposes and premises. Users are worried about whether AI can distinguish these differences.

It should be noted that what is introduced here is a summary of some of the posts that could be confirmed. It is not a survey showing the ratio of pros and cons across all social media, nor has it been confirmed that modding and preservation activities have been uniformly restricted.

The Anxiety of Those Being Copied Cannot Be Overlooked

On the other hand, understanding this movement solely through criticism of the restrictions is insufficient.

Game Developer reports that concerns about AI imitation are spreading among indie game developers, leading Digital Sun Games to cancel the public playtest of 'ReVamp'.

Publishing a work in progress has the meaning of having people play it, gathering opinions, and increasing those waiting for its release. If the publication is felt to potentially trigger imitation, developers face a difficult decision between promotion and defense.

What is important here is not just whether a highly complete copy can actually be made.

The anxiety that "something similar might be published first" itself can make developers hesitate to distribute demos or share progress. If opportunities for publication decrease, so do opportunities for players to encounter new works.

From this perspective, it is meaningful for AI providers to show a policy of curbing imitation. However, if the mechanism stops developers who should be supported, it creates another problem.

For Japanese Indie Developers, Both Positions Overlap

For Japanese indie developers or small teams, this issue becomes even more relatable.

When creating their own games, they want AI to help implement and fix programs. On the other hand, they feel anxious about others creating similar works in a short time based on published demos or promotional videos.

The same developer can be on the side seeking free use of AI and on the side seeking to curb imitation.

This is why it is difficult to resolve with a simple choice of "regulate or not."

What is needed is an explanation of which parts were judged problematic and a procedure to correct misjudgments. It is also important that when it is communicated that the materials are self-made or that there is permission from the rights holder, it is appropriately reflected in the judgment.

If AI simply refuses the task, users are left without clues for improvement. To be trusted as a development support tool, not only the precision of stopping but also the precision of allowing legitimate creation is questioned.

From "Can It Be Made" to "What to Make"

Regarding this change, the scope of application also needs confirmation.

The Verge's report confirms Anthropic's explanation of changes to the prompts for Claude Code on Opus 5.5. The media has inquired whether updates have been made or will be made to other Claude system prompts.

Therefore, it cannot be concluded from this report that the same restrictions are uniformly applied to all Claude products or models.

Also, the fact that AI accepted a request does not mean that the rights issues of the resulting product are resolved. Conversely, just because AI refused does not mean that the creation is illegal.

What will be questioned in future AI game creation is not just the high ability to reproduce. What was referenced, what was thought of independently, and what materials were used. Being able to explain that process also holds value.

Creating play influenced by favorite works while making something unique. How to design AI that helps with that.

The two sentences added by Anthropic do not complete that answer. It is a change that highlights the difficulty of simultaneously protecting creators and not stopping their hands.

Sources and References

1. The Verge: Reference for changes to Claude Code, Osmani's explanation, and the unconfirmed scope of application to other models.
   https://www.theverge.com/ai-artificial-intelligence/1024632/anthropic-claude-ai-video-game-clonestheverge.com

2. Addy Osmani's X post and public display of replies: Explanation of prompt changes, consistency with learning data, and reactions regarding mods, preservation, and permitted use. Below is an external viewing page for X posts. SNS statements are summarized in the text.
   https://xitter.cf/addyosmani/status/2116213240363729013teapawt

3. Public display of chasm's post: Report of discovering new rules and raising questions about the impact on game remakes and mod creation. Below is an external viewing page.
   https://xitter.cf/chasmmmmmmmmmmm/status/2116062991686721968teapawt

4. Game Developer: Reference for confirming Anthropic's policy and the background of developers concerned about AI imitation, such as the cancellation of the public playtest of 'ReVamp'.
   https://www.gamedeveloper.com/programming/anthropic-appears-to-have-ordered-ai-tool-claude-to-stop-remaking-video-gamesgamedeveloper.com