Anthropic updates Claude to block users from cloning existing games

New restrictions on Claude

Anthropic has introduced a specific update to its Claude Opus 5.5 model to prevent the tool from being used to clone existing video games. The change was identified in the model's internal rules and was later confirmed by a member of the company's technical staff. This move is intended to steer the AI away from recreating specific titles that belong to other developers.

The new instructions explicitly tell the AI to build original games, games within a specific genre, or games based on a user's own designs and assets. It specifically forbids the creation of a recreation of a specific existing game, even if the user tries to rename the project or call it something original. This policy is designed to ensure the tool is not used to infringe on intellectual property.

According to Addy Osmani, a member of Anthropic's technical staff, the company added two sentences to the prompt to enforce this boundary. Osmani noted that using Claude to infringe on intellectual property is against the company's usage policy. The update aims to maintain a balance where original creative work remains supported while direct copies are blocked.

The rise of AI slop

This policy shift comes after weeks of the Steam platform being inundated with what many call AI slop. These are low-effort, vibe-coded games that are often heavy imitations or direct copies of existing titles. The influx of these games has caused significant disruption and outcry within the broader game development community.

The problem became particularly visible during the most recent Steam Next Fest. During this event, the number of showcased demos doubled compared to the previous event, with AI-assisted games accounting for most of that increase. Many developers have reported that their games and demos are being flagrantly copied with the help of AI tools.

The situation has become severe enough that several studios have chosen to cancel planned public demos of their upcoming titles. These developers fear that if they release a demo, the code could be quickly reverse engineered. Once reverse engineered, the code could be used by others to create AI-generated clones of their work.

Technical capabilities and risks

Claude Opus 5.5 has demonstrated a high level of proficiency in reverse engineering code, which has created various risks for the industry. These capabilities have been linked to the rapid decompiling of old console games and even the jailbreaking of the PS5. Such technical abilities make it easier for bad actors to lift codebases wholesale rather than just riffing on ideas.

While many of the games being cloned are relatively simple in design, the core issue remains the unauthorized use of existing code. Games like Vampire Survivors and Balatro are built on elementary concepts that have naturally spawned imitators. However, Anthropic's new rule is specifically aimed at stopping the wholesale lifting of a codebase rather than the imitation of a game's genre or mechanics.

The goal of the new prompt is to ensure that the AI continues to function for legitimate developers who are building from their own designs. Anthropic stated they will continue to iterate on these rules to find the right balance. They have encouraged users to report instances where the AI might incorrectly turn down a project that is truly original.

Steam's stance on AI tools

Valve, the company that operates the Steam platform, has admitted that the store is facing a problem with AI slop. However, the company has expressed reluctance to solve the problem by banning AI coding tools entirely. Valve's leadership stated they do not want to be gatekeepers who tell developers which specific tools they should or should not use.

Instead of a ban, Valve suggested that developers focus on making their games harder to copy by adding unique features. They advised creators to keep growing their games by adding personality and creativity that stems from their own unique perspectives. This approach puts the responsibility on the developers to protect their work through creative depth rather than platform-wide restrictions.

As the industry navigates this transition, the tension between AI efficiency and intellectual property remains a central theme. Anthropic's move to hardcode these boundaries into Claude represents one way for AI providers to mitigate the risks of their tools being used for plagiarism. The industry will likely continue to see a mix of AI-assisted creation and efforts to safeguard original content.