Rohan Paul

Anthropic is cracking down on video game clones. Looks like the rule arrived through a server-side flag, “tengu_heron_brook”, so it reaches every session without an update and has no off switch. The instruction still permits original games, games in a genre and projects built from a user’s own designs and assets. Exact rule text: “Do build original games, games in a genre, and games from the user’s own designs and assets. Don’t build a recreation of a specific existing game, even if it is renamed or called original.” To enforce the above rule it at scale, Anthropic can reuse the classifier stack it built for weapons jailbreaks: small models trained on a written rule list that read each request with its output, plus cheap probes on Claude’s internal activations that flag intent to copy a specific title. Research tools go further, comparing code even after renaming, matching runtime behaviour and sprites, and measuring “distance” between game designs. i.e. measuring distance between them with concept vectors. The hard part is legal: copyright protects a game’s look and feel, not its mechanics, so every detector has to separate “same genre” from “same game,” and that is where false refusals will come from.