Claude Code blocks game clones. Developers say it is blocking their work too.

Claude Code is being told not to recreate someone else's specific video game. That is a reasonable line to draw. The awkward bit is that developers now report being stopped at the door to their own workbench too.

Anthropic has confirmed the narrow part. In a statement cited by Game Developer on 30 October, Anthropic engineer Addy Osmani said the company added two sentences to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game. The reported instruction says original games, games in a genre, and games made from the user's own designs and assets should still be built. A renamed copy of a particular existing game should not. Anthropic ties this to its rule against infringing third-party intellectual property.

That is more precise than “Claude has banned game development.” It is also narrower than the claim that Anthropic has formally banned all reverse engineering. The problem is that a narrow rule on paper can behave like a wide one in practice.

The developer at the barrier

One public report is unusually concrete. In an open Claude Code GitHub issue filed on 30 October, the operator of Conflict Simulations LLC says Opus 5.5 began refusing to continue work on commissioned digital board-wargame modules after a server-delivered game instruction appeared. The business says it writes original code for private clients, checks that they own the physical game and uses materials the clients supply. It asks Anthropic for a way to verify this kind of professional work and for clearer notice when server-side instructions change.

There is an important qualification in the issue itself: the reporter corrected an earlier description and says this work is not publisher-licensed. Owning a copy of a game does not automatically settle every copyright or licensing question about making a digital version. We cannot independently determine whether every requested module was permitted, or reproduce the exact refusal from that report. But the account does show the practical problem: a coding assistant may be making a rights-sensitive decision from a short prompt, with little room for a developer to explain the project.

Another GitHub report, filed on 24 October, says a classifier blocked a patch to the author's own board-game project. That issue is brief and tagged as needing reproduction, so it is a signal, not a verified diagnosis of the new game rule. Tom's Hardware also reports user complaints about refusals during anti-cheat work, security reviews and ordinary troubleshooting, including analysis of a Windows crash dump with a reverse-engineering skill. Those accounts deserve investigation, but they do not prove one universal ban or establish that the same safeguard caused every refusal.

Why the distinction matters

There are good reasons to stop a tool from pumping out close copies of somebody else's game. A small studio can spend years on a distinctive idea, then watch a lookalike appear in a weekend. An AI coding tool that accelerates that copying has a real effect on creators, not just a theoretical IP problem.

But game development is also full of work that looks similar to copying from a distance. A studio may port its own title, rebuild a lost prototype, inspect a crash dump, test an anti-cheat system, or study how a genre convention works before making something original. A blunt refusal at any mention of “recreate,” “reverse engineer” or a familiar game mechanic can halt legitimate work while doing little to stop a determined copier who simply rephrases the request.

Anthropic's published Usage Policy prohibits violating third-party IP rights. It also says a real-time block is not, by itself, a finding that the user broke the policy. That distinction matters here. A refusal can be a model or safeguard error, not a legal judgment on the project. Anthropic has said it will iterate and invited examples when Claude turns down work that belongs to the user.

What would make the rule workable?

The useful test is not whether Claude can say “no” to an obvious clone. It is whether it can say “yes” to a clearly original project, and handle the grey areas without pretending it is a copyright court. Developers need a visible explanation of what triggered a refusal, a practical route to contest mistakes, and advance notice when server-delivered instructions alter work they rely on. Anthropic needs to show that the false-positive rate falls without reopening the door to straightforward copying.

For now, the confirmed change is a targeted Claude Code instruction against recreating a specific existing game. The wider pattern of blocked reverse-engineering and troubleshooting work is reported, not yet established as a blanket policy. The cartoon version is a robot guard stopping everyone at the gate. The real engineering job is harder: let the people with their own plans and tools through, while turning away the actual copy machine.