Anthropic blocks Claude Code from cloning existing games

Anthropic has pushed a rule to Claude Code that stops the AI coding tool from recreating existing video games, with the change landing on October 27 under the internal name "tengu_heron_brook."

X user Chasm first spotted the update. The rule instructs the model to build "original games, games in a genre, and games from the user's own designs and assets," but explicitly not "a recreation of a specific existing game, even if it is renamed or called original." According to Chasm, users cannot switch the rule off.

Anthropic member of technical staff Addy Osmani confirmed the change on X. The company had "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," Osmani wrote.

He added: "Original games, genre games, and games built from your own designs and assets are meant to keep working. We'll continue to iterate to ensure we're striking the right balance."

What does the rule actually change?

For studios and solo developers, two operational consequences stand out.

• Legitimate use should keep working. Genre projects and games built from a user's own designs and assets remain fully supported, per Osmani's statement.
• Clones are off by default and off permanently. The rule cannot be toggled off by the user, according to Chasm's findings — a hard-coded constraint rather than a preference.

The mechanism itself is modest: two sentences added to the prompt running on Opus 5.5. Anthropic is framing this as prompt-level steering, not a model retrain or a separate classifier. That distinction matters for anyone betting on enforcement strength.

Can the restriction be bypassed?

As Game Developer points out, effectiveness is uncertain. Safety restrictions on AI models can be bypassed with carefully worded prompts, and Claude has no obvious way to verify whether a user actually owns the game design or assets they supply. A developer who claims ownership of someone else's design — or simply describes it in original-sounding terms — may still get a functional clone out of the tool.

So the rule raises the friction floor without building a wall. It signals Anthropic's policy position on IP infringement; it does not guarantee compliance.

Why now?

The update arrives amid rising developer anxiety over AI-assisted game cloning. Digital Sun recently delayed a public playtest of its game ReVamp specifically over fears that its demo could be fed into an AI tool and reproduced. Valve, meanwhile, has faced pressure to explain how it will stop AI-made clones from crowding out original games on Steam.

Those two pressures — studios withholding playable builds, and platform holders sitting on an unclear enforcement posture — form the commercial backdrop for Anthropic's move. A toolmaker acting before a platform maker is a notable sequencing.

For teams using Claude Code in production pipelines, the practical guidance is simple: keep your own design docs and assets attached to prompts, and expect genre-adjacent work to pass while named-game recreations fail. For rights holders, the rule is a first acknowledgment from a major AI vendor that game cloning is a policy problem, not a hypothetical one.

Watch whether Valve responds with its own AI-clone policy for Steam — and whether OpenAI and Google match Anthropic's prompt-level restriction in their coding tools.