Anthropic has changed how Claude Code handles game projects. Around 27 October 2026, a new instruction began reaching Claude Code from Anthropic's servers under a flag named tengu_heron_brook, according to X user chasm, who spotted it and posted about it on 30 October (Japan time). Users cannot switch it off in their settings, GameBusiness.jp reports. As shared by chasm, the rule tells Claude not to build a recreation of a specific existing game, even if it is renamed or called original. Anthropic has not published the exact wording.
What Anthropic confirmed: Addy Osmani of Anthropic said on X that the team added two sentences to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game, because using Claude to infringe intellectual property breaks Anthropic's usage policy. He said original games, genre games and games built from a user's own designs and assets should keep working. He also said Anthropic will keep adjusting the balance and asked users to tell him if Claude refuses work on something they own the rights to.
Reported, but not confirmed: Tom's Hardware counts a wave of GitHub issues about refusals during game development, when copying a game mechanic and when patching a tabletop-style game. Users on Reddit and XDA Developers also say Claude refused harmless reverse-engineering, such as analysing a Windows crash dump, removing junkware from a Samsung TV, or using the Reverse Engineer Anything (REA) tool. Some game developers say anti-cheat work and security reviews of their own games were blocked. Anthropic has not said whether these refusals come from the new rule, and Tom's Hardware says the timing may be a coincidence.
Why now: in late September, people posted Opus 5.5 recreations of hit games such as Terraria and Rocket League, GameBusiness.jp reports. Some indie developers have since received emails threatening to release an AI clone of their game first unless they pay. In Japan, Alliance Arts dropped the public demo of PUKEY GODDESS SHOT TRICK on 16 October, citing fears of AI copies, and the developer of Project Solaris stopped twice-weekly progress posts. On social media, critics accuse Anthropic of a double standard, since AI training itself faces copyright disputes.
