Anthropic tightens Claude Code rules to block recreations of existing games

Anthropic narrows what Claude Code will build

Anthropic has changed the instructions behind Claude Code, its AI coding assistant, so that the tool will not recreate specific existing video games. The Verge, citing an earlier report from Tom's Hardware, says the system prompt that guides Claude Code was updated to block recreations of particular titles. Building games is still allowed — the restriction targets close replicas of named, commercially released works.

What the new prompt says

According to The Verge, a user on X posted a screenshot last week that appeared to show two fresh sentences in Claude Code's system prompt. The substance: the assistant should build original games, games within a genre, and games from a user's own designs and assets, and it should not build a recreation of a specific existing game, even when the request claims the project is original. The Verge notes the screenshot is unverified, and Anthropic did not confirm that exact wording.

Anthropic's explanation

Addy Osmani of Anthropic addressed the change publicly in a reply on X. He did not vouch for the precise language in the screenshot, but confirmed the company had added two sentences to Claude Code's prompt on Opus 5.5. His stated rationale: using Claude to infringe intellectual property goes against Anthropic's usage policy, and the new guidance is intended to steer the tool away from copying somebody else's specific game. He said original games, genre games and games built from a user's own designs and assets are meant to keep working, that Anthropic will continue tuning where the line sits, and that users should send him cases where Claude wrongly refuses work that is genuinely theirs.

Why now

The Verge frames the change against the retro gaming scene, which it describes as flooded with AI-generated decompilations, recompilations and ports of classic titles. Coding agents like Claude Code can produce playable software in a single session, and hobbyists have used them to resurrect old games or build near-identical replicas. That activity carries legal risk: freshly written code can still infringe when it reproduces protected assets, characters, art or audio, even though copyright generally does not cover abstract game mechanics or ideas.

Open questions

The Verge says it asked Anthropic whether other Claude system prompts, such as the one behind the standard chatbot, have been or will be updated the same way; no answer is reported. It also remains unclear how the model will distinguish a genre homage from a recreation in practice, how often borderline requests will be refused, and whether the policy rests on prompt guidance alone or will be backed by technical enforcement. Osmani's invitation to report false refusals suggests Anthropic already expects some misfires.

Why it matters

The move shows AI companies starting to embed copyright policy directly into system prompts — the hidden instructions that shape an assistant's default behaviour — rather than relying solely on after-the-fact enforcement of usage rules. As coding agents become able to generate entire working applications on request, the question of what they will and will not build becomes a policy decision with real legal weight. It also raises a transparency issue: users learned of this change through a screenshot and a follow-up comment rather than an announcement. For developers, the practical boundary matters most — games in a genre are allowed, recreations of specific titles are not, and the space between those two is exactly where disputes over cloned mechanics, art styles and level design will land. Anthropic says it will keep adjusting the balance, and how it calibrates that line will be worth watching for anyone using AI tools in creative software.