Anthropic blocks game cloning but accidentally restricts real developers

New restrictions on Claude Code

Anthropic recently updated the internal instructions for Claude Code, its command-line programming tool. The company added two specific sentences designed to prevent the Opus 5.5 model from recreating existing video games. This move was intended to stop users from violating intellectual property by making clones of popular titles.

The update was first noticed by a user on X and later confirmed by Addy Osmani, a member of the technical team at Anthropic. The new rules are meant to be specific, allowing users to create games based on original ideas or general genres. However, the tool will now outright refuse any request to recreate a specific, existing title.

This policy change comes in response to growing concerns about AI-assisted game cloning. During the recent Steam Next Fest, the number of participants doubled, and AI-assisted games made up a significant portion of that growth. Many of these entries were noted as being barely disguised copies of existing work.

The impact on independent studios

Small game studios have already reported significant issues regarding AI theft. Some developers have shared stories of extortion, where others threatened to turn their demos into full games before the original creators could finish them. One studio even canceled a public demo for their game to prevent its code from being harvested by AI models.

Anthropic's new filter aims to solve this problem by enforcing usage policies. By blocking the recreation of specific titles, the company hopes to protect the work of these smaller creators. The goal is to ensure that the tool is not used as a shortcut to bypass the hard work of original game design and development.

Despite these intentions, the implementation of the filter has caused friction. The company is trying to balance the need for copyright protection with the utility of the tool for legitimate programmers. This balance is proving difficult as the AI struggles to distinguish between a clone and a legitimate request for help.

Overreach and developer backlash

The new safety measures have led to significant backlash from the developer community. Many users report that Claude Code is now blocking them in the middle of their own original projects. These developers are being denied help with simple tasks, such as fixing a specific game mechanic or performing a routine security audit on their own software.

Beyond game development, the filter is affecting other types of technical work. Some users have reported that the model refuses mundane reverse-engineering tasks, such as analyzing crash files from a Windows PC. Others have tried to use the tool to clean up bloatware on a Samsung TV and were met with the same refusals.

The situation has led some developers to express frustration over the irony of the situation. Critics point out that these models were trained by vacuuming up large portions of the internet without permission. They argue it is hypocritical for the creators to become strict sticklers for copyright only when their customers are the ones attempting to copy content.

Alternative paths for users

Because of these restrictions, some tinkerers are beginning to look for alternatives. A number of users have reported migrating to local models or Chinese models that do not impose these specific types of questions. These models allow for more freedom in reverse-engineering and complex coding tasks without the same level of automated gatekeeping.

Valve, the company behind Steam, has taken a different approach to the issue of AI cloning. Rather than acting as a police force for the platform, Valve advises developers to make their games harder to copy. They suggest that putting more unique personality into a game makes it significantly more difficult for an AI to replicate effectively.

While Valve's advice is practical, many developers find it difficult to execute in a competitive market. The struggle between protecting intellectual property and maintaining a functional tool for developers continues to be a primary tension in the AI space. For now, the community is waiting to see how Anthropic refines these filters to be less intrusive.

Next steps for Anthropic

Anthropic has acknowledged the issues caused by the overreaching filter and has promised to adjust the settings. The company wants to find a way to block malicious cloning without hindering the productivity of legitimate software engineers. They are looking for a way to make the tool more discerning about what constitutes a violation of policy.

To help resolve the immediate problems, Addy Osmani has invited developers who were blocked on their own projects to contact him directly. This direct line of communication is intended to help identify specific cases where the filter is malfunctioning. By gathering this feedback, the technical team hopes to refine the instructions for the Opus 5.5 model.

The situation serves as a case study in the challenges of AI safety and copyright. As companies move to protect creators from automated theft, they risk creating significant hurdles for the very people they aim to support. The next few weeks will be critical in determining whether Anthropic can successfully tune these boundaries.