Anthropic adds anti‑cloning prompt to Claude as AI game copying surges
Anthropic updated Claude’s system prompt to block recreations of existing games, but the move may not curb the broader AI‑driven cloning boom.
Source: Ars TechnicaAnthropic has inserted a new system prompt into its Claude models that explicitly forbids recreating specific existing video games. The change, announced late October, aims to keep the company on the right side of copyright law, yet the broader tide of AI‑assisted game cloning shows little sign of receding.
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.Original games, genre games, and games built from your own designs and assets are meant to keep working.We’ll continue to iterate to ensure we’re striking the right balance. In the meantime, if Claude turns down something that’s yours, send it my way.— Addy Osmani (@addyosmani) October 30, 2026
What new rule did Anthropic add to Claude?
According to the Ars Technica report, a JSON rule named "tengu_heron_brook" was added to the Claude Code prompt around October 27. The rule instructs the model to "Do build original games, games in a genre, and games from the user’s own designs and assets. Don’t build a recreation of a specific existing game, even if it is renamed or called original." XDA confirmed the rule appears in version 2.1.314 of Claude Code, released on October 30, and noted that Claude Opus 5.5 carries a similar restriction in its system‑prompt bundle. Anthropic software engineer Addy Osmani later tweeted that the two‑sentence addition was meant to steer Claude away from infringing IP while still allowing truly original creations【https://x.com/addyosmani/status/2116213240363729013】.
Why did Anthropic introduce the anti‑cloning prompt?
The motivation is two‑fold. First, it provides a defensive layer against potential copyright lawsuits that could arise if Claude were used to generate near‑identical copies of protected games. Anthropic has faced similar legal scrutiny over image‑generation models that reproduce copyrighted artwork. Second, the rule aligns Claude’s usage policy with the company’s public stance that using its models to infringe IP is prohibited. By embedding the restriction in the model’s prompt, Anthropic hopes to reduce the number of inadvertent violations that could expose the firm to liability.
How effective is the rule likely to be?
The article argues the rule will have limited impact on the overall cloning ecosystem. Developers seeking to replicate a game can simply switch to other frontier models that have not implemented comparable safeguards. Open‑weight alternatives, which are often cheaper and only marginally less capable, remain readily available. Moreover, the enforcement relies on Claude interpreting a very nuanced legal distinction—what counts as a "specific existing game" versus a "genre‑style" or "transformative mashup"—a task that even sophisticated LLMs struggle with. Early reports from Anthropic’s GitHub issue tracker indicate the new prompt sometimes blocks legitimate development workflows, suggesting the filter may be over‑inclusive.
What does the wave of AI‑driven game copying look like?
Since the release of powerful code‑generation assistants, a flood of projects has emerged that decompile, port, or remix classic titles. Communities such as decomp.dev, once devoted to painstaking manual reverse‑engineering, have shut down because AI tools enable rapid, token‑driven porting that skews community metrics. The report cites examples of AI‑generated mashups—Spider‑Man swinging over Arkham Knight’s Gotham—shown in a YouTube video that demonstrates how a model can stitch together assets from two unrelated franchises【https://www.youtube.com/watch?v=PLRjyX7GykI】. Another video highlights a fully playable Spider‑Man prototype built on top of a different game engine【https://www.youtube.com/watch?v=9bpjXcUyaHQ】. While these creations are technically impressive, they raise open questions about copyright, especially when recognizable characters or assets are included.
How can Grafx users work with these developments?
Creators who want to experiment with AI‑assisted game design can do so safely using Grafy’s dedicated tools. The Game Studio app lets you turn a textual brief into a playable prototype, but it respects the same IP constraints you would apply manually—avoid uploading copyrighted assets unless you own them. For developers who need to write or refactor code, Grafy Code offers a sandboxed environment where you can test Claude‑style prompts against your own codebase without risking accidental infringement. If you are unsure whether a concept crosses the line, the Reader app can ingest the Ars Technica article (or any PDF) and let you ask the model specific questions about the legal nuances. Finally, the Chat interface lets you compare Claude’s behavior side‑by‑side with other models that may or may not have similar restrictions, giving you a clearer picture of what each system will allow.
Source: Ars Technica, "Claude’s new rules against game cloning won’t stop the AI copying boom" (Nov 2 2026) – https://arstechnica.com/ai/2026/11/claudes-new-rules-against-game-cloning-wont-stop-the-ai-copying-boom/
Frequently asked questions
Does Claude now refuse to generate any game that resembles an existing title?
Claude’s updated prompt blocks exact recreations of a specific game, even if the user renames it. It still permits genre‑style games or original projects that use the user’s own assets. Edge cases—such as mashups or games that share mechanics but not assets—remain ambiguous and may be handled inconsistently.
Can I still use Claude to create a brand‑new game concept?
Yes. The prompt explicitly allows building original games, genre‑based games, and games that rely on the user’s own designs and assets. As long as the output does not copy protected code, art, or story elements from a known title, Claude should comply.
Will other AI coding models adopt similar anti‑cloning safeguards?
There is no guarantee. While Anthropic’s move reflects growing legal pressure, competitors have not announced comparable system‑prompt restrictions. Developers seeking to avoid IP risk may need to evaluate each model’s policy individually or rely on open‑weight models that lack built‑in safeguards.
Source: Ars Technica — Claude’s new rules against game cloning won’t stop the AI copying boom
More news
Personal AI Agent Got Pentagon Data
An Accenture contractor’s personal Claude agent accessed a Pentagon system for two months, raising national security concerns.
Anthropic's Mythos-class model allegedly leaked on Hugging Face
An alleged leak of Anthropic's Mythos-class model has surfaced on Hugging Face, sparking security concerns and community discussion. Read the details.
Claude Duet Shows How Personal‑Assistant Agents Can Be Fast, Organized and Credential‑Free
Claude Duet, Anthropic’s new personal‑assistant harness, delivers snappy replies.