Anthropic has given the programming assistant Claude Code a new instruction: no recreation of a specific existing game, „even if it is renamed or passed off as an original“. Anthropic confirmed the change for the Opus 5.5 model to the US trade outlet Game Developer. Original game ideas, games within a genre, and games based on your own designs remain explicitly allowed.
What did Anthropic change in Claude Code?
A spokesperson referred Game Developer to a statement by Addy Osmani from Anthropic's technical team. According to Osmani, two additional sentences have been in the prompt that Claude Code works with under Opus 5.5 since the change. Anthropic cites its own usage policy as the reason, which prohibits using Claude to infringe third parties' intellectual property rights.
Previously, a user on X had discovered the rule under the internal name „tengu_heron_brook“ in the instructions that Claude Code received from the server around October 27. According to his report, the instruction cannot be turned off. The change is missing from the release notes of Claude Code up to and including version 2.1.316.
What remains allowed, and where does the rule fall short?
According to Osmani, original games, games within a genre, and games based on your own designs and assets should continue to work. The line is drawn at the specific model: Claude Code will still program a racing game, but not a racing game „like Need for Speed“, reports Kotaku.
What Claude Code still builds for games
Rule for Claude Code with Opus 5.5, active since around October 27, 2026 according to user reports
- Original game ideas
- Games within a genre, such as a racing game
- Games from your own designs and assets
- Recreation of a specific existing game
- The same recreation under a new name
- Switch to turn off the rule
According to a bug report on GitHub, contract work on games owned by clients also fails
Sources: Game Developer, Kotaku, GitHub
A developer who programs digital versions of strategy board games for clients, which those clients own, describes in the bug report on GitHub that Claude Code has been refusing such jobs since the end of October, on both the Max and Team plans. A response from Anthropic is still outstanding in the report, which has been open since October 30, as of November 2.
How should the rule be assessed?
Game Developer sees the instruction against the backdrop of a wave of AI clones. Digital Sun Games, for example, cancelled a public playtest for ReVamp out of concern that someone could copy the demo with an AI tool. At the same time, the trade outlet doubts its effectiveness: protective rules in AI tools can be circumvented, Claude Code cannot verify who owns a game, and freely downloadable models have no such block.
What is new is less the prohibition than the place where Anthropic draws the line: the usage policy already prohibited infringements before, but now the block takes effect directly in the tool, without a switch and without an entry in the release notes. Because the instruction comes from the server, the rule also affects Claude Code with Opus 5.5 in Germany; Anthropic has not named any regional exception.
What this means for you
If you are planning a game with Claude Code, it is best to describe your own idea: genre, rules, characters, and graphics that you designed yourself. If the prompt names a well-known title as a model, Claude Code refuses the task according to the reports so far. How a game is created with Claude from the idea to a playable result is shown by the live test with Claude as a game developer in the Claude course on TutKit.
For contract work on an existing game, even with the permission of the rights holder, Anthropic has so far provided no exception. The bug report proposes unlocking verified professionals, similar to the existing program for verified security researchers. It is therefore best not to rely firmly on Claude Code for such projects and to report such refusals to Anthropic: Osmani has promised to keep fine-tuning the rule.
Game Developer published the report on October 30, 2026 and subsequently added Anthropic's statement.
Sources
