Claude Code now refuses to clone specific games. Anthropic confirms two sentences did it.
A staffer confirmed the rule is real and small. Developers say the refusals reach well past games, and nobody outside Anthropic can yet say where the line sits.
Anthropic has told Claude Code not to rebuild a specific existing video game, and an Anthropic staffer has now confirmed it in public. The whole change, by his account, is two sentences added to the prompt that Claude Code runs on Opus 5.5.
My read: it’s a small edit with a big backdrop. Indie studios have spent weeks worrying that someone will feed their demo to an AI tool and ship a copy first. Here is how the pieces fit together, in order.
October 6: cyber access narrows
Tom’s Hardware reports that Anthropic launched its Cyber Verification Program on October 6. Advanced cybersecurity capabilities in its frontier models are limited to entities covered by Project Glasswing, with the change effective immediately. That matters later, because some of the refusals developers now complain about may come from this layer and not from the game rule. Tom’s Hardware says the timing of the two “may be a coincidence.”
October 27: two sentences go in
According to Game Developer, an X user called Chasm spotted a new entry in Claude Code’s rules, pushed sometime on October 27 under the flag “tengu_heron_brook.” It reads: “Do build original games, games in a genre, and games from the user’s own designs and assets. Do not build a recreation of a specific existing game, even if it is renamed or called original.” Chasm reported there is no way to switch it off.
The week of refusals
Tom’s Hardware says Claude Code’s GitHub tracker has filled with issues about refusals during ordinary game work: one halfway through development, one over replicating a game mechanic, one over patching a tabletop-style game. Users also claim Claude has refused to analyze a Windows crash dump and to clear junkware off a Samsung TV, per the outlet’s reporting and XDA’s. Tom’s Hardware presents these as user claims, says some of the GitHub reporters may have bad intentions, and adds that at least one is a developer with a real track record.
Tom’s Hardware also cites purported game developers who say Claude blocked anti-cheat work and security reviews on their own games.
Anthropic’s answer
PC Gamer quotes Addy Osmani, a member of Anthropic’s technical staff, replying to Chasm: “We 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 IP is against our usage policy.” He added that original games, genre games and games built from your own design “are meant to keep working,” and that if Claude turns down something that’s yours, “send it my way.”
Osmani’s reply covers the game rule. It says nothing about reverse engineering in general, the Windows crash dump or the TV.
| Claim | Who says it | Status |
|---|---|---|
| A two-sentence rule on game recreation was added to Claude Code’s prompt on Opus 5.5 | Osmani, Anthropic, via PC Gamer | Confirmed by Anthropic staff |
| The rule was pushed on October 27 and can’t be disabled | Chasm, via Game Developer | One user’s report |
| Claude refuses broader reverse-engineering and security work | Users, via Tom’s Hardware and XDA | Unverified, cause unknown |
Why studios care
PC Gamer says several studios have cancelled public demos for fear their code will be reverse engineered into a clone. Game Developer names Digital Sun Games, which cancelled a public playtest of its upcoming game ReVamp. PC Gamer adds that the latest Steam Next Fest had twice the demos of the previous one, with AI-assisted games accounting for most of the increase, and that Valve’s Tom Giardino told developers to keep adding “features, personality and creativity” that come from their own perspective.
My read: this rule treats a symptom. Game Developer points out that safety limits on AI tools can be talked around, that Claude has no obvious way to tell whether a user owns a game idea, and that cloners can switch to open-weight models with no such rule. A prompt line can only turn away the laziest copy. It also hands honest developers a new way to get stuck in the middle of a project.
The part I’d watch is the line between “a game in the genre of Balatro” and “Balatro again.” Osmani says Anthropic will “continue to iterate” on that balance. If the GitHub issues keep piling up, that iteration will happen in public, and our Sonnet 5.5 release post covers the stricter cybersecurity safeguards that came with Opus 5.5.
Note on sources
I read the Game Developer, PC Gamer and Tom’s Hardware articles in full. I couldn’t open Osmani’s original X post, so his quote comes from PC Gamer. The XDA report on the Samsung TV is cited only as Tom’s Hardware links to it, and I didn’t read it myself. Anthropic’s usage policy page exists, but I’m not relying on it for any detail here.