Anthropic curbs video game clones in Claude Code: a useful rule, but easy to get around

By Game Developer · October 30, 2026.

Anthropic has updated Claude Code, its AI programming tool, so that it won't recreate other people's specific video games. Addy Osmani, a member of Anthropic's technical staff, confirmed it in a post on X. According to him, two sentences were added 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 intellectual property violates the company's usage policy.

The rule was spotted by X user Chasm, who found it in Claude Code's "rules" published on October 27 under the label "tengu_heron_brook." The text reads: build original games, games in a genre, or games based on the user's own designs and assets; do not build a recreation of a specific existing game, even if it is renamed or presented as original. According to Chasm, there is no way to turn it off.

Osmani clarified that original games, genre games and those made with one's own designs and assets should still work. He added that they will keep iterating to "find the balance" and asked to be told if Claude refuses something legitimate.

The backdrop is unease among independent developers. According to Game Developer, fear has spread in recent weeks of fast AI-made clones reaching the market before the original game. Digital Sun Games canceled its public playtest of ReVamp for fear that someone would feed the demo into an AI tool and release their own version. Developers have also pressed Valve to explain how it will prevent clones from displacing their games on Steam. Valve gave an unusual explanation to developers in South Korea this week.

**Our take.** It is a reasonable move and, above all, a signal of direction: a model provider acknowledges it cannot look the other way when its tool makes it easy to copy others' work. But it shouldn't be overstated. A two-sentence instruction in a prompt is a soft barrier. The outlet itself notes that researchers and journalists have documented that generative AI safety limits can be bypassed with rhetorical tricks. It is also unclear how the model will decide whether a user owns an idea or a set of assets. The line between "genre game" and "clone" is blurry even for lawyers.

There is a second problem: competition. As several users pointed out in replies to Chasm, if Claude becomes inconvenient, anyone wanting to clone can turn to open-weight models, which have no such restrictions. More broadly, the open frontier is rising fast, and that is good news for cost, control and sovereignty. But it means abuse prevention cannot rest on a single provider: it will have to rely on platforms (Steam), intellectual property law and after-the-fact detection.

In the short term, the message to small studios is mixed. The barrier to copying is falling, eroding the advantage of those who show their work early: hence the canceled demos. In the long term, the same capability that enables cloning lets a two-person team make an ambitious, original game. That is the part we want to protect. The sensible approach is for rules to target copying, not the tool, and to be judged by their false positives: if Claude refuses what is yours, the balance has not been struck.

A note of rigor: the rule is a prompt change, not a technical guarantee, and the source provides no metrics on its effectiveness. Until there are, it is a statement of intent with real effect against the least determined copycats, nothing more.