📅 October 30, 2026 • Source: Game Developer
Anthropic has added a server-side rule to Claude Code that instructs the Opus 5.5 model not to recreate existing games. The rule reads: «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 change lands amid an ongoing debate about a wave of AI-generated game clones that reuse the mechanics and visual identity of independent releases.
How it works. The prompt is delivered server-side under the internal identifier tengu_heron_brook. It was noticed around October 27 by a user on X under the handle chasm and was independently found inside the binary of Claude Code v2.1.314 by XDA. There is no user-facing switch to disable the rule locally, so behavior is consistent across installations and cannot be opted out of through configuration. A member of technical staff at Anthropic, Addy Osmani, confirmed the change in a post on X: «We added two sentences to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game because using Claude to infringe IP is against our usage policy.» He added that original games, games in a genre and projects built from the user's own assets continue to work, and invited developers to write to him directly if Claude refuses something of theirs.
IMPACT For studios, the rule lowers the risk that Claude Code is used to produce close copies of shipped games, a concern that became concrete when the studio Digital Sun canceled the public playtest of its game ReVamp citing fear of cloning. The same context shows the friction the rule creates: a developer of licensed tabletop wargames filed GitHub issue #106167 stating that the rule blocks his paid client work, and some developers consider the wording too broad, since genre conventions and specific recreations are covered for the model by one instruction. Teams working on licensed or derivative material are left a documented path: Anthropic points to direct contact when legitimate requests are refused.
The episode is a concrete example of content policy being enforced inside developer tooling at the prompt level rather than only in a terms-of-service document. Tool behavior changed without a local setting, and the discovery came from users inspecting the binary rather than from a changelog. The practical takeaway for teams is that server-side prompts can alter tool behavior between versions, and releases of coding agents may now carry policy changes that deserve the same review attention as functional updates.
Sources: gamedeveloper.com and cryptobriefing.com
Continue the conversation: https://t.me/AIAutomationArchitecture · Bluesky