Claude Code's new rule against cloning games is a signal of intent, not real protection for game developers

Anthropic has quietly updated Claude Code, its AI coding agent, so it will not recreate specific existing video games. Game Developer reports the change. An X user named Chasm spotted a new rule pushed around October 27 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.” According to Chasm, users cannot switch it off.

Anthropic confirmed the change. A spokesperson pointed to a statement by Addy Osmani, a member of Anthropic's technical staff. He said the company added two sentences to Claude Code's prompt on Opus 5.5 because using Claude to infringe intellectual property (IP) breaks its usage policy. Original games, genre games and games built from a user's own assets “are meant to keep working,” he wrote. He said Anthropic will keep adjusting the balance and invited users whose own projects get refused to contact him.

The context explains the urgency. Over the past few months the industry has seen a wave of AI-built remakes of classic and even unreleased games, some made for fun and some for profit. Fear of fast AI clones that reach the market before the original has spread among independent developers. Digital Sun Games is one of several studios that cancelled a public playtest of an upcoming game, in its case ReVamp, for fear that someone would feed the demo into an AI tool. Developers have also pressed Valve to explain how Steam will stop clones from displacing their games. Valve gave a rare explanation to developers in South Korea this week.

The limits are obvious, and Game Developer lists them. Researchers and journalists have repeatedly shown that safety limits on generative AI can be bypassed with rhetorical tricks. It is unclear how the model can tell whether a user actually owns a design or its assets. Above all, frustrated users can switch to open-weight models with no such restriction. Open-weight models are those whose weights anyone can download and run.

Our reading: this is a governance signal, not a technical fix. A frontier lab is saying that its agent's capabilities come with defined permitted uses. That fits a pattern we have flagged before. As agents grow more capable, the competitive edge shifts from what the model can do to what it is allowed to do, and who is accountable. The move does have a cost. A two-line prompt rule will produce false refusals for legitimate developers and will barely slow down determined cloners. It also highlights an uncomfortable tension in the open-weight ecosystem, which we generally see as a force for democratization. The same openness that lowers costs also removes the guardrails. The lasting answer will not come from a system prompt. It will come from platforms like Steam, from provenance and authorship tools, and from IP enforcement updated for a world where copying takes hours rather than months. In the long run, AI that lets anyone build a game is excellent news for creativity. In the short run, the industry has to make sure it does not become a machine for copying other people's work.