Anthropic Adds Rule Barring Claude Code From Recreating Specific Existing Games

Anthropic has quietly told its coding assistant to stop cloning other people's video games. The rule surfaced around Oct. 27 and has already produced complaints from developers who say it is catching their own work.

The instruction sits in Claude Code's rules under the internal label "tengu_heron_brook." It 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 rule is delivered server-side, and users have reported no setting to switch it off.

What Anthropic says

Addy Osmani, a member of Anthropic's technical staff, confirmed the change in a reply to the X user Chasm, who spotted it. "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," Osmani wrote.

He said original games, genre games and games built from a user's own design are meant to keep working. "We'll continue to iterate to ensure that we're striking the right balance," he wrote. "In the meantime, if Claude turns down something that's yours, send it my way."

Anthropic has not published a formal announcement.

The problem it targets

Game developers have spent weeks complaining about AI-assisted copying. Several studios have canceled planned public demos because they fear the code will be reverse engineered and turned into clones.

Some developers have also been hit by extortion scams. The scammers threaten to turn a studio's demo into a full game before the original can ship.

The most recent Steam Next Fest showed the scale of the flood. The number of showcased demos doubled from the previous event, and AI-assisted games accounted for most of the increase. Valve has said it is weighing a response.

Opus 5.5 is also credited with rapid decompiling of old console games and with the jailbreak of the PS5. Open-source tools such as Reverse Engineer Anything (REA) have been used to port major franchises, including Grand Theft Auto V, Call of Duty and Halo, into web browsers without authorization.

Developers say legitimate work is getting blocked

The flip side is a growing pile of refusals. Claude Code's GitHub page has seen an influx of issues from users blocked midway through development, while replicating a game mechanic, and while patching a tabletop-style game. At least one of those reports comes from a developer with a verifiable track record.

Others say Claude balked at analyzing a Windows crash dump, clearing junkware off a Samsung TV, or using REA at all. Some self-described game developers say the model blocked anti-cheat work and security reviews on their own games, claiming its safeguards were triggered.

Some could come from people with bad intentions, and Tom's Hardware noted as much.

The core complaint from developers is that Claude cannot reliably tell someone cloning a commercial game from someone working on their own code. Anthropic's own wording that it is still iterating to find "the right balance" concedes the line is not yet clean.

An open question on cyber restrictions

The timing overlaps with another Anthropic change. On Oct. 6, the company introduced its Cyber Verification Program, which limits access to its frontier models' advanced cybersecurity capabilities to entities covered by Project Glasswing.

Tom's Hardware speculated that the same ruleset could be behind the security-related refusals. Anthropic has not said whether the game rule and the cyber restrictions are connected.

The property rights angle

The underlying principle is not exotic. Copying someone else's game and slapping a new name on it is infringement, and a company is entitled to refuse to help with it. Developers who spent years building a title have a legitimate interest in not seeing it cloned by a chatbot in an afternoon.

But a hidden, non-removable prompt rule that misfires on a developer's own code is a real cost to people doing nothing wrong. Anthropic has put the burden on those users to contact Osmani directly and ask for a fix.

What comes next

Anthropic has not said how it will measure false positives or whether it will narrow the rule. Valve has not announced what, if anything, it will do about AI-generated clones on Steam. The next test is whether developers still report refusals on their own projects once Anthropic's promised adjustments land.