Skip to content
Neural Daily

ToolsNews

Claude Code will no longer clone existing games

🇭🇺 Olvasd el magyarul →

Anthropic added two sentences to Claude Code's Opus 5.5 system prompt: build original and genre games, but not a recreation of a specific existing game, even if it is renamed. The rule arrived from Anthropic's servers without a version change.

By: Mikola Máté Patrik · · 3 min read

Claude logo with the orange starburst mark on a cream background
Image: Anthropic · Press kit · Official social image of the Claude pricing page · Source

What happened

Around October 27, an X user spotted a new rule in Claude Code's prompt: "Do build original games, games in a genre, and games from the user's own designs and assets. Do not build a recreation of a specific existing game, even if it is renamed or called original." The post passed two million views.

Addy Osmani of Anthropic confirmed the change (Dexerto). He said two sentences were added to Claude Code's prompt on Opus 5.5 because using Claude to infringe IP is against the usage policy. Original games, genre games and games built from your own designs and assets should keep working, and he asked users to send him cases where Claude wrongly refuses their own work.

Why now

According to Engadget, AI-assisted decompiled AAA titles such as Halo: Combat Evolved, Call of Duty: Modern Warfare 2 and GTA V have recently appeared as browser-playable fan projects. Several developers have also complained about AI-made clones of their games on Steam. Anthropic also agreed to a $1.5 billion copyright settlement last year, so caution is understandable.

The more interesting part for developers

The rule did not ship with a new model or client release. According to GitHub issues cited by sysnote, Claude Code reads a server-controlled prompt slot (tengu_heron_brook) and inserts its text verbatim into the system prompt. In July, an experiment used the same slot to tell the model not to call subagents unless the user asked, which overrode one project's CLAUDE.md rule.

In other words, the behavior of an agent with shell access can change overnight while the version number stays the same. That is not necessarily malicious, but it is worth knowing if you build reproducible agentic workflows.

How strict is it?

Early tests suggest it is narrower than the headlines: the model refused to build a Pokémon clone but helped with a decompilation project. Meanwhile, Claude Code's GitHub has seen complaints about refusals on ordinary game work. It is also unclear how the model could know whether assets really belong to the user, and people set on cloning can simply switch to a local open-weight model.

If you build games with Claude Code, say clearly in your prompt that the design or assets are your own. If you hit a wrong refusal, report it to Anthropic.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles