Anthropic adds prompt instructions to block game clones in Claude Code

A new guardrail for game dev

Anthropic has updated its Claude Code tool to steer users away from rebuilding existing video games. As reported by shattered.io, the company added two specific sentences to the Claude Code prompt on the Opus 5.5 model. These instructions are designed to push the model away from recreating specific games that belong to other people.

The update follows a period of speculation after users discovered a new internal flag. A user on X shared that the tool had picked up a new instruction regarding game creation. This discovery led to reports that Anthropic is actively monitoring and restricting how its AI coding agent handles game development requests.

Addy Osmani, a member of Anthropic's technical staff, confirmed the change in an interview with Game Developer. He explained that the goal is to ensure the tool is not used to infringe on intellectual property. This move turns an abstract usage policy into a specific behavior for a high-traffic use case.

How the restriction works

The restriction is implemented as a prompt addition rather than a hard filter or a retrained model. A prompt instruction provides a nudge to the model's behavior using plain language. This allows the AI to understand the boundary of what it should and should not build based on the user's request.

The delivery method is significant because the rule arrives from Anthropic's servers via a flag called tengu_heron_brook. This means there is no user-facing setting or toggle to switch the rule off. Users cannot disable it in a configuration file or tick a box to declare their project as legitimate.

The wording of the new instructions makes a clear distinction between genres and specific titles. The model is encouraged to help users build original games or games within a specific genre, such as a farming sim. However, it is instructed to decline requests to build a faithful recreation of a specific existing game, even if the user tries to rename it.

Impact on developers and hobbyists

The rule creates a complex situation for different types of creators. Original studios and small indie developers may benefit from a tool that makes it harder to produce low-effort clones. By raising the cost of copying, the tool may help protect the market share of unique, original titles.

Conversely, hobbyists who use AI to learn by rebuilding classic games may find their workflow interrupted. Many people use faithful remakes as a common teaching project to understand game logic. If the model identifies these requests as recreations of specific titles, it will refuse to assist with the code.

Modders and preservationists also face uncertainty regarding how the model interprets their work. Some projects involve rebuilding source code from shipped binaries for preservation or platform ports. It remains unclear if the model will treat these specific types of technical tasks as recreations or as permissible technical work.

Legal and market implications

Anthropic's move is driven by the need to manage legal exposure and protect its brand. While copyright law generally does not protect game mechanics or ideas, it does protect specific expressions like code and art. By blocking clones, Anthropic avoids being seen as a provider of tools for efficient infringement.

The restriction is a vendor-level choice that manages the company's own risk. It signals to rights holders that Anthropic does not want its products used for unauthorized copies. However, it does not stop people from writing code by hand or using other tools that do not have these specific guardrails.

For users who want to bypass these restrictions, open-weight models remain an option. Because the rule is a hosted prompt, users running models locally face no such instructions. This creates a divide between managed AI services with strict policies and local models that offer more freedom for high-risk projects.