Anthropic changes Claude rules for building games
Claude should now allow game references for genre and mechanics while redirecting direct requests to copy an existing title.
Anthropic has changed the way Claude responds to requests involving existing video games. Instead of treating every reference to a named game as a possible cloning request, Claude should now allow developers to describe the mechanics and genre they want, while still refusing to reproduce the original game itself.
The change is intended to separate legitimate game development from direct copying. It follows complaints that the earlier restriction was blocking developers who were using a famous game only as shorthand for the kind of system they wanted to build.
What Anthropic changed
The update was outlined by Anthropic technical staff member Addy Osmani in an X post cited by Tom's Hardware. Osmani said a short rule added to the Opus 5.5 prompt had been stopping Claude Code from handling some ordinary game projects. The revised behaviour is designed to let users name a game when describing mechanics, but offer to build something original in the same genre if they ask for a copy.
That distinction matters. A request such as building an original platformer with mechanics similar to a familiar series should be treated differently from a request to recreate a specific game's characters, levels, art or code. Anthropic's new approach appears to preserve that line while reducing false positives for developers who are discussing influences rather than asking for a replica.
The change is tied to Claude Code version 2.1.317, according to Osmani's post. It does not amount to a blanket permission to reproduce copyrighted games, and it does not make a generated project legally safe simply because Claude produced it. The user remains responsible for the material they ask the tool to create and the way they distribute the result.
Why the restriction appeared
AI coding tools have become increasingly capable at producing playable prototypes, but those same capabilities can be used to imitate existing software at speed. Game decompilation and reverse engineering have traditionally required specialist knowledge and substantial time. New AI tools can lower that barrier, creating a more obvious route to infringing copies, unofficial ports and projects built from another developer's work.
Tom's Hardware also reported concerns about bad actors using AI-assisted reverse engineering against small and independent developers. The pressure on Anthropic is therefore two-sided: the company needs to prevent Claude from becoming a convenient game-cloning tool, while keeping it useful for studios and hobbyists who are building original work.
Anthropic has described the revised direction as an attempt to protect developers' work without blocking legitimate creative and technical use. The result is a narrower rule rather than a complete removal of safeguards.
What developers can expect
For practical use, the safest route is to describe the desired genre, systems and player experience in original terms. Asking Claude to create a new game with turn-based combat, resource management or a particular camera perspective should fit the intended use case. Asking it to reproduce a named game's exact levels, characters, interface or code should still trigger a refusal or a redirect towards an original alternative.
The distinction may not always be perfectly clear. Genres borrow from one another, and many modern games share established mechanics. A model that is too cautious can make ordinary design conversations frustrating; a model that is too permissive can help produce near-identical copies. Anthropic's updated prompt is an attempt to move the boundary towards context-aware judgement rather than relying on a simple list of prohibited names.
The move also highlights a wider issue for AI-assisted development. As generated code becomes more useful, developers need clearer provenance and review practices, especially when an idea is described through a famous reference point. Claude can help turn a design brief into a prototype, but it cannot decide whether the finished project crosses a legal or ethical line in every jurisdiction.
Our opinion
Anthropic's adjustment is sensible because a named reference is not automatically a cloning request. Developers regularly use existing games to explain pacing, control schemes or genre conventions, and blocking every such conversation makes an AI coding tool less useful without necessarily preventing bad-faith copying. The more defensible boundary is the one Anthropic now describes: allow discussion of mechanics and genre, refuse requests for a direct reproduction, and make the user do the work of turning the inspiration into something genuinely original.
The policy will only be convincing if Claude applies that distinction consistently. False positives will remain frustrating, while false negatives could put smaller developers at greater risk. Version 2.1.317 is therefore a measured policy correction, not an all-clear for AI-generated game cloning.