Anthropic has updated its Claude artificial intelligence models with new instructions barring them from recreating specific, existing video games, targeting intellectual property infringement across its coding tools.
The restriction surfaced in version 2.1.314 of the Claude Code binary, which launched on October 30. Embedded under a configuration rule, the directive instructs Claude Opus 5.5 to permit original titles, broad genres, and user-provided assets, but explicitly commands it not to build recreations of existing games, regardless of whether a project is renamed or presented as original work. Anthropic software engineer Addy Osmani confirmed the addition on social media, stating that using Claude to infringe on intellectual property violates the company’s usage policies.
The guardrail was implemented amid a surge in automated decompilations, retro ports, and rapid AI-generated game clones. That influx recently led decompilation community decomp.dev to shut down, citing a wave of low-effort ports that disrupted its collaborative project model. However, developers have already reported friction with Anthropic’s new prompt, filing issues on the Claude Code GitHub repository when the filter blocked legitimate development tasks.
According to Ars Technica’s analysis, the guardrail primarily protects Anthropic from potential copyright liability rather than halting the wider trend of AI game copying. Developers seeking to recreate existing titles can still bypass the restriction by utilizing competing commercial systems or unrestricted open-weight models. The prompt also leaves the AI model with the difficult task of distinguishing between protected IP, derivative mashups, and standard genre conventions.
Source: arstechnica.com
