Anthropic Restricts Claude Code to Prevent Video Game Cloning

Anthropic implemented server-side restrictions on its Claude Code tool, specifically for the Opus 5.5 model, to prevent the AI from recreating existing video games. The update, identified by the configuration flag 'tengu_heron_brook', adds two sentences to the model's prompt to prohibit the reproduction of specific existing titles, even if they are renamed, while still supporting original game development and genre-based projects.

Addy Osmani of Anthropic stated the move is intended to prevent intellectual property infringement, which violates the company's usage policy. This action follows a trend of AI-assisted decompilation of major titles such as Grand Theft Auto V and Halo: Combat Evolved. The restrictions coincide with Anthropic's Cyber Verification Program, which limits full cybersecurity capabilities to approved entities.

Software developers have reported that these guardrails extend beyond game cloning to include legitimate reverse-engineering tasks and security checks. Some users noted that switching to the Opus 4.8 model bypasses these blocks. Independent developers, including Conflict Simulations LLC, have reported false positives where the AI refused commissioned work for board game modules. While Anthropic has invited developers to report incorrect refusals, it has not yet established a formal verification process for authorized commissioned work.