Anthropic has quietly ordered Claude Code to stop helping rebuild existing games, and the instruction is exactly two sentences long. An X user known as chasm first spotted on October 30 that the coding agent had received a new rule around October 27, pushed from Anthropic's servers under a configuration field named tengu_heron_brook, with no setting to turn it off. The rule reads: "Do build original games, games in a genre, and games from the user's own designs and assets. Don't build a recreation of a specific existing game, even if it is renamed or called original."
Anthropic confirmed the change the same day. Addy Osmani, a member of the company's technical staff, said 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," adding that original games, genre games and games built from a user's own designs and assets "are meant to keep working." A spokesperson referred Game Developer to the statement. Notably, there is no changelog entry or blog post — users learned of the rule from one person reading his own config file.
XDA Developers located the exact instruction in version 2.1.314 of the Claude Code binary, along with a separate delivery route tied to Opus 5.5. In XDA's tests, Opus 5.5 refused to recreate a specific Nintendo title but agreed to help with a decompilation project and outlined a static recompilation — suggesting the model can still tell reverse engineering apart from recreation, at least sometimes.
Not every user sees that line drawn. A GitHub issue opened October 30 by a developer whose company builds digital board-wargame modules for customers who own the physical games says Claude Code began refusing that paid work; the issue was still open on November 1, and other developers have reported refusals on wholly original projects. XDA said it had not reproduced those refusals. The gap between the rule's narrow wording and its enforcement is the familiar prompt-policy problem: users cannot audit what prompt text shapes their tools, and Anthropic has not said which accounts receive which delivery route.
Why now? The preceding months produced a wave of AI-accelerated game rewrites: community ports running World of Warcraft, Call of Duty and Grand Theft Auto in browsers, crossover mods dropping Mario into Elden Ring and Spider-Man into Batman's city, and a PlayStation 5 jailbreak running the Xbox exclusive Halo 2 and Nintendo's Zelda on Sony hardware. After extensive Grand Theft Auto 6 leaks, Take-Two subpoenaed Microsoft and Discord hunting for the source. During a recent Steam Next Fest, low-quality clones of newly released demos flooded in — one studio, Alliance Arts, cancelled its public demo outright, and users demonstrated a playable Pokemon clone assembled from a single prompt in under two hours.
Critics have called this the game industry's "Napster moment": a technology that collapses the cost of copying is now trying to close the box it opened. There is an obvious tension — frontier labs trained on the open web now police IP violations downstream — and Anthropic itself is defending a copyright suit from Elvis Presley's rights holders over song lyrics. A two-sentence prompt is unlikely to stop anyone determined to clone a game with a different model, but it signals how quickly labs convert legal exposure into prompt text, and how little visibility users have when it happens.
Comments (0)
Log in to join the discussion
Log InNo comments yet