Claude Code now refuses to rebuild specific existing games, even if renamed

Anthropic has changed Claude Code so that, on Opus 5.5, it steers away from recreating someone else’s specific game. The company says the change is two sentences added to the prompt, and that original and genre games should keep working.

Two sentences in the prompt, found by a user

According to Game Developer, the rule was spotted by an X user named Chasm, who found it in Claude Code’s rules under the label tengu_heron_brook. Chasm said it was pushed around October 27. Anthropic has not confirmed that date, according to the report.

As posted by Chasm, the rule allows building original games, games in a genre, and games made from the user’s own designs and assets. It bars building a recreation of a specific existing game, even if the result is renamed or described as original. Chasm also reported that users cannot switch the rule off.

An Anthropic spokesperson pointed reporters to a statement by team member Addy Osmani. He said the company added “two sentences” to Claude Code’s prompt because “using Claude to infringe IP is against our usage policy.” Osmani added that original, genre and own-asset games are meant to keep working, and that Anthropic will keep iterating to find the right balance. He also invited developers to report mistaken refusals: “if Claude turns down something that’s yours, send it my way.”

Who is affected and what remains unclear

The change applies to Claude Code users working with Opus 5.5. Reporting does not say whether other models in the tool are affected, and no price, plan or regional difference is involved. For a developer building a platformer, a roguelike or a card game of their own, nothing should change. For someone who asks Claude Code to rebuild a named commercial game, the tool is now meant to decline, and renaming the project is not supposed to get around that.

Several points are not established. The rule text comes from a user’s post, not from an Anthropic document. It is unclear how the software would judge whether a user owns the idea or assets behind a game. Game Developer also notes that the rule may only deter low-effort copycats, since jailbreak methods have defeated AI limits before, and that commenters pointed to open-weight models without such limits as an alternative for cloners.

A copyright rule enforced through the prompt, not the law

The step is notable for how it works. Osmani framed it as enforcing an existing usage policy on intellectual property, and Anthropic chose a prompt-level instruction rather than a technical block. A prompt steer is cheap to ship and quick to adjust, which fits his promise of iteration, but it is also a soft control that depends on the model reading intent correctly.

The pressure behind it is visible among developers. Game Developer reports that Digital Sun Games, maker of the upcoming game ReVamp, canceled public playtests over fears that someone could feed a demo into an AI tool and clone it, and links the trend to Moonlighter’s developer delaying its playtest. Developers have also pressed Valve over clones on Steam. PC Gamer ties the Claude change to weeks of Steam being flooded with AI-made games, and a summary of its coverage says Valve does not want to police which tools developers use.

The scale of AI use on the store is large. Tom’s Hardware reported that about one in five Steam games released in 2025 disclosed generative AI use, with 7,818 titles disclosing it, up nearly 700 percent year on year and about 7 percent of the whole library. Disclosure is not the same as cloning, and many of those games are original, but the figures show why developers feel exposed. Anthropic’s own safeguard covers only its tools: a clone built with another model is untouched, and Valve announced no new rules in the coverage reviewed.

The episode also shows the other side of AI-assisted building. Our guide to vibe coding describes how prompting alone can now produce working apps, which is what lets a whole game be reproduced quickly from a demo. Whether two sentences can hold that line, and how many legitimate projects get caught, will only become clear from the cases developers send Osmani.