Anthropic has introduced a subtle but significant constraint to its Claude Code system prompt, explicitly instructing the model to avoid recreating specific existing video games. Discovered by AI decompilation enthusiast "chasm" and verified by XDA in version 2.1.314, the rule is buried in a JSON file under the identifier "tengu_heron_brook." The directive allows for original games and genre-based projects but blocks direct recreations of existing titles, even if renamed. This move signals Anthropic's growing anxiety over intellectual property liability as AI coding tools fuel a massive surge in game porting and cloning.

Liability Over Capability

The primary driver behind this prompt engineering appears to be legal risk mitigation rather than technical limitation. Anthropic software engineer Addy Osmani confirmed the change on social media, stating that "using Claude to infringe IP is against our usage policy." Osmani emphasized that the update is meant to steer the model away from generating code that copies someone else’s specific game, while still permitting games built from user designs and assets. This distinction is critical for Anthropic, which faces increasing scrutiny over how its models interact with copyrighted works, similar to the legal headaches seen in the image generation sector.

The Decompilation Crisis

The timing of this update coincides with a community backlash against AI-assisted game decompilation. Last month, decomp.dev, a long-standing community hub for reverse engineering, announced its shutdown. Maintainer Luke Street argued that AI tools had broken the community aspect of decompilation, turning a brutal, time-consuming craft into a "hollow numbers game" where newcomers race to port games without contributing to the underlying effort. The new Claude prompt seems designed to align Anthropic with traditional developer sentiment, but it arrives after the genie is already out of the bottle, with thousands of AI-powered ports already flooding the ecosystem.

Why This Won't Stop the Boom

Despite Anthropic's efforts, the restriction is unlikely to slow the broader trend of AI game cloning. Developers can easily switch to other frontier models that have not implemented similar restrictions, or move to cheaper open-weight models that lack these specific guardrails. The capability gap between Claude and its competitors is not wide enough to force compliance, especially for hobbyists and indie developers who prioritize speed over strict IP adherence. Furthermore, the prompt creates ambiguous legal gray areas: does merging Mario into Elden Ring count as a recreation of a specific existing game or a transformative work? Asking an LLM to reliably adjudicate copyright fair use is a tall order, and courts themselves struggle with these definitions.

Key Takeaways

  • Anthropic added a system prompt rule in Claude Code v2.1.314 to block specific game recreations, identified as "tengu_heron_brook."
  • The change is driven by IP liability concerns, with Addy Osmani confirming that IP infringement violates usage policy.
  • The decompilation community is fracturing, with decomp.dev shutting down due to AI-assisted vibe-coded ports overwhelming human-centric efforts.
  • Competitors and open-weight models offer easy workarounds, limiting the real-world impact of Anthropic's restriction.
  • The prompt fails to clearly define transformative works, such as game mashups, leaving developers in a legal gray area.

The Bottom Line

Anthropic is playing legal defense, not technical innovation. By adding a prompt guardrail instead of a capability barrier, they’ve created a compliance theater that competitors will ignore and developers will bypass. The tsunami of AI clones isn’t stopped by a two-sentence instruction; it’s stopped by market saturation and quality control, neither of which Anthropic controls.