Anthropic’s anti-cloning rules can't stop the AI game-copying boom

Anthropic recently updated its system prompt to stop Claude models from re-creating specific existing video games. While this move helps shield the company from liability, it will do little to slow the massive wave of AI-assisted game porting sweeping the industry. The new rule, spotted in the Claude Code binary and confirmed by engineers, explicitly tells the model not to build recreations of specific games even if renamed.

But eager creators can simply switch to competing frontier models or cheaper open-weight alternatives that lack these restrictions. This flood of AI-generated copies has already upended traditional human-led communities. Retro game decompilation projects have exploded in volume, turning a painstaking craft into a hollow numbers game and causing long-running hubs like decomp.dev to shut down.

Indie developers are now afraid to show early work publicly for fear of rapid cloning. Our read is that while Anthropic's guardrails manage the company's legal exposure around intellectual property infringement, they treat the symptom rather than the systemic capability now built into modern coding tools. Enforcing this boundary also forces an AI to solve legal questions that confound human courts.

The prompt allows original genre games but blocks specific clones, a gray area that grows murkier when users ask models to mash up disparate intellectual properties like Spider-Man and Batman.

Asking an automated system to reliably draw the line between a transformative work and an illegal copy is a tall order.

Watch whether other AI labs follow Anthropic's lead or if open-weight models simply absorb the entire copying boom.