Anthropic adds rules to stop Claude from cloning existing games
The AI company is implementing new guardrails to prevent its coding assistant from recreating specific copyrighted games.
The short version
- Anthropic added two sentences to Claude Code's prompt on Opus 5.5 to steer the model away from recreating specific games.
- The AI is still permitted to build original games, games within a specific genre, or games based on user-provided assets.
- The policy change aims to prevent users from using the AI to infringe on the intellectual property of existing software.
- Critics note the rule focuses on output rather than the data used to train the models in the first place.
Why it matters
This shift signals that AI companies are beginning to face pressure from software giants to police specific outputs. For developers, it means the 'vibe-coding' of clones will face more friction as models become more aware of brand boundaries.
New guardrails for Claude
Anthropic has introduced a new rule to prevent its AI-powered coding assistant, Claude, from recreating specific existing games. This move comes as headlines have frequently highlighted users attempting to create "vibe-coded" clones of popular software and big-ticket game titles. The company is attempting to address the growing issue of AI-assisted intellectual property (IP) infringement.
As reported by gizmodo.com, the update specifically targets the wholesale recreation of existing games. The AI is now instructed that it can build original games, games within a specific genre, or games based on a user's own designs and assets. However, it is explicitly forbidden from building a recreation of a specific existing game, even if the user attempts to rename it.
Addy Osmani, the Engineering and Evangelism Leader at Anthropic, confirmed the change via a social media post. He stated that the company added two sentences to the Claude Code prompt on Opus 5.5 to steer the model away from these specific recreations. Osmani noted that using Claude to infringe on IP is against the company's usage policy.
The logic of the restriction
The way Anthropic drafted this rule is significant because it uses a "do everything except" approach. Instead of starting with a blanket rule of what is allowed, the prompt defines a broad permission and then creates specific exceptions for what is prohibited. This method requires the company to anticipate and list every potential "naughty" use to keep the AI in check.
This approach can be more difficult to manage than a "do nothing unless permitted" framework. By allowing the AI to build games generally, the company creates a situation where things might inevitably slip through the cracks. It places the burden on the developers to constantly update the prompt with new prohibitions as new games or software products become targets for cloning.
Because the rule is framed this way, it relies on the AI's ability to recognize what constitutes a "specific existing game." If the AI cannot perfectly identify a clone, the guardrail may fail. This creates a technical challenge in ensuring the model understands the nuances of what makes a game a unique piece of intellectual property versus a generic genre piece.
Who is affected by the change
The immediate impact of this change is felt by developers and hobbyists who use Claude to generate code for software projects. While the AI remains a powerful tool for building original software, users will find it harder to prompt the model into generating code that mimics a known commercial product. This is a direct response to the trend of using AI to quickly generate functional clones of existing hits.
Software companies with large legal teams, such as Adobe or Microsoft, are also stakeholders in this policy shift. While the current rule only mentions games, there is a possibility that these restrictions could expand to other types of software. If people can use AI to create knockoff versions of expensive industry-standard subscription software, these companies may demand even stricter protections.
There is also a noted irony regarding how these rules are applied. While Anthropic is now policing the output of the AI to protect specific companies, critics point out that the models were trained on vast amounts of work from the same creators. This creates a tension between the company's desire to protect current IP and its previous actions regarding the data used to train the models.
Potential for loopholes
The current restriction may be susceptible to semantic trickery by savvy users. For example, it is unclear how the model will react to a request to build a game that has not been released yet, such as a hypothetical sequel to a popular franchise. If the AI recognizes the brand but not a specific released game, the current wording might not catch the request.
Furthermore, the rule does not explicitly address knockoffs of entire intellectual properties that don't mirror one specific game. A user might ask for a game that captures the
Sources and links
- PostOctober 30, 2026 x.com
- Reportinggizmodo.com: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by gizmodo.com · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
