Anthropic confirms Claude Code's game-cloning restriction as developers report false refusals

Anthropic has confirmed a change intended to stop Claude Code from recreating someone else’s specific game. The dispute now centers on whether that instruction also catches legitimate development work.

In an October 30 statement, Anthropic technical staff member Addy Osmani said the company added two sentences to Claude Code’s prompt on Opus 5.5 because using Claude to infringe intellectual property violates its usage policy. Game Developer reports that an Anthropic spokesperson referred the publication to that statement.

“Original games, genre games, and games built from your own designs and assets are meant to keep working,” Osmani wrote. He said the company would keep adjusting the balance and invited users to send him cases where Claude rejects work that belongs to them.

The confirmed rule has a specific target

Game Developer traces the controversy to a user who reported finding an instruction around October 27: build original games and games within a genre, but do not recreate a particular existing game, even under another name. Anthropic’s subsequent explanation confirms the intended restriction on specific-game recreation.

That matters for developers using a coding assistant to prototype familiar mechanics. A genre resemblance and a request to copy someone else’s particular game are different categories in the company’s stated intent. The statement does not establish that Claude consistently draws that boundary correctly.

Users report a wider effect

Tom’s Hardware describes users reporting refusals involving their own game work, game mechanics and non-malicious reverse-engineering tasks. Those accounts are evidence of a dispute about how the assistant behaves; they do not confirm a blanket Anthropic policy prohibiting all reverse engineering, modding or original game development.

The practical issue is overblocking: a developer can own the design and assets yet still receive a refusal. Anthropic acknowledges that possibility by asking for examples. Its public response promises further iteration, without establishing how common such failures are or when particular cases will be resolved.