Claude Code rule blocks recreating existing games

What the new Claude Code rule says

Claude Code is Anthropic's AI coding tool. You describe a change in a terminal or editor, and it writes and edits the code for you. Claude Code on Opus 5.5 still builds original, genre, and self-designed games, but it now turns down requests to rebuild a specific existing game.

A two-sentence rule pushed from Anthropic's servers

The rule surfaced on October 30, when X user Chasm posted that Claude Code had picked up new instructions around October 27. The rule arrives from Anthropic's servers under a flag called tengu_heron_brook, and users have no setting to switch it off. Chasm's post also quotes the full text, which runs just two sentences.

The rule as posted by Chasm

Sentence	What it tells Claude Code
First	Go ahead with original games, genre games, and games based on the user's own designs or art
Second	Say no to a copy of one particular existing game, even under a new name or an "original" label

The second sentence closes the obvious loophole. Giving a familiar game a new name, or calling it your own, doesn't change the answer.

Addy Osmani confirms the change and cites the usage policy

Addy Osmani, a member of Anthropic's technical staff, replied to Chasm the same day, saying the two sentences went into Claude Code's system prompt (the standing instructions it reads before your request) on Opus 5.5. Osmani said Anthropic added the two sentences because using Claude to infringe IP breaks its usage policy, and that the team will keep adjusting where the line sits. The policy itself bans using Anthropic's products to infringe or misappropriate a third party's intellectual property.

An Anthropic spokesperson pointed Game Developer to Osmani's post when asked for comment. Anthropic hasn't published a separate announcement, so that post is the company's only explanation so far.

A board wargame studio says paid client work stopped

Conflict Simulations LLC says Claude Code stopped its paid work turning clients' own board wargames into digital modules, and support would not make an exception. Board wargames are tabletop games that simulate battles or whole wars on a map. The studio's owner, Ray Weiss, argues that a rule aimed at games as a category can't tell piracy apart from a commission for someone who already owns the game.

Three requests filed on GitHub

Weiss opened a GitHub issue (a public bug report or feature request) on the Claude Code repository on October 30. The business builds personalized digital versions of board wargames for individual clients who own the physical games.

The code is written from scratch, clients supply the materials themselves, and each client shows proof of ownership first. Sessions that had done this work for months now refuse to go on, according to Weiss.

Weiss asked Anthropic to limit the rule to real infringement, to offer a verification path for ownership-checked work, and to log changes like this in the changelog. Weiss compared the second request to the program Anthropic already runs for vetted security researchers.

Later that day Weiss corrected one word. The first draft called the work "licensed," but it isn't licensed by publishers. The builds are for each client's own use, never sold or shared, and the studio won't build a module for a game whose publisher already sells a digital edition.

A follow-up comment adds a detail about timing. The work kept going after the rule reportedly shipped and only stopped once Weiss asked Claude to check whether the rule applied to it. Clearing that session didn't help in the same repositories.

Another user replied that the memory files or Git history were probably carrying the refusal forward. Weiss answered that they would try to clean those up.

Support offered no exemption, and the changelog is silent

Weiss also contacted Anthropic support from two accounts. Each time, the Fin support bot replied that there's no exemption or verification path for commercial users. Both tickets later reached a person, and one was handed to a specialist team with no timeline.

Anthropic's support bot said the policy applies to every customer without exception, and the Claude Code changelog has no entry for the rule. As of November 2, the changelog runs through version 2.1.316 and says nothing about games. The Next Web reported the same gap.

Not everyone in the thread sided with Weiss. One commenter said that unless the clients hold the rights, the work is no different from any other reverse-engineered port (a rebuild made by taking apart the original software).

Which games can you still build with Claude Code?

Anything that is not a rebuild of a specific existing game is still fair game, and if your own project gets blocked, Osmani has asked to hear about it. That covers new ideas, games built around a familiar genre, and projects made from your own designs and art. Only a copy of one particular game is off the table.

Original, genre, and self-designed games are allowed

The first sentence of the rule names three kinds of game Claude Code should keep building. Osmani said the same three are meant to keep working.

• Original games you came up with yourself
• Games in an established genre, such as a platformer or a puzzle game
• Games built from your own designs and assets

Renaming a well-known game or calling it original does not get a recreation past the rule. The second sentence covers both moves directly.

Game Developer doubts the rule will stop determined copycats. It noted that researchers have repeatedly shown AI safety limits can be talked around, and that it's unclear how software would know whether someone owns a game idea or its assets. Game studios are on both sides of this, too: Epic Games has moved its own development toward AI (Epic Games delays Disney shooter, goes AI first).

What to do if Claude Code blocks your own game

Developers whose own work gets turned down can flag it to Osmani, who asked for those cases while Anthropic tunes the rule. The request came in the same post that confirmed the change.

Support is a weaker route for now. The board wargame studio asked through two accounts and was told no exemption exists. Right now the public channels are Osmani's post on X and the issue tracker on the Claude Code GitHub repository.

Server-side rules change Claude Code without a version bump

Two ways Claude Code's behavior can change

Change	Listed in the changelog?
A new app version	Yes (2.1.316 was the latest on November 2)
A server-side instruction, like this rule	No, and there's no setting to turn it off

Because the rule comes from the server, skips the changelog, and has no off switch, what Claude Code will build can change while your version number stays the same. Osmani has said Anthropic will keep adjusting the balance, so the wording could change again without notice.

Version updates are easier to trace. Auto mode, for example, became the default startup mode starting with version 2.1.283, and that kind of change can be pinned to a release (Claude Code Auto Mode Is Now the Default Setting). This rule has no such marker.

The wargame studio's timeline adds another wrinkle. Work kept going after the rule reportedly arrived and stopped only after Weiss asked Claude about it. If session memory or Git history really does carry a refusal forward, two people on the same version could hit the rule at different times.

So if Claude Code suddenly stops on game work, check the server-side rule before blaming your version. That gap is exactly why the studio asked for changelog entries.