Claude Code Now Blocks Recreating Existing Video Games

• Anthropic added two sentences to Claude Code's Opus 5.5 prompt blocking recreations of existing games, confirmed October 30

• The rule bars renaming a clone as original but still allows genre games and your own designs, per Anthropic's wording

• There is no setting to turn the rule off, and some developers report it blocking original commissioned projects too

• If you build games with Claude Code, document originality before you start or risk an unexplained refusal today

Claude Code will not rebuild your favorite game under a different name anymore. Anthropic confirmed on October 30 that it quietly added a rule steering Opus 5.5 away from recreating existing titles, and there is no toggle to switch it off. The change landed without a changelog entry, and plenty of developers found out the hard way.

What Did Anthropic Actually Change?

Developers first spotted the new behavior around October 27, when Claude Code started refusing prompts that asked it to rebuild a specific existing game, even a decades-old arcade title with no active copyright enforcement behind it. The refusals looked like a model quirk at first, the kind of thing you shrug off and retry with different wording.

They were not a quirk.

Addy Osmani, who joined Anthropic to work on Claude Code's developer experience, confirmed the cause once reporters pushed for an answer: two sentences were added to the Claude Code system prompt, scoped to Opus 5.5. Multiple outlets covering the change, including Game Developer and Dexerto, converged on close to the same wording. Build original games, games in a genre, and games from your own designs and assets. Don't build a recreation of a specific existing game, even if it's renamed or called original.

That last clause is the part that stings. It's not just "don't copy Tetris." It's "don't copy Tetris and call it Falling Blocks." Anthropic's public Usage Policy already prohibits using its products to infringe someone else's IP, so this prompt addition enforces a rule that technically already existed on paper. It just wasn't enforced at the model-behavior layer before October.

I'd read the timing as the real story here, more than the two sentences themselves. PCGamer tied the change to weeks of Steam getting flooded with AI-assisted game clones, cheap reskins of popular titles shipped fast enough to ride a trend before anyone noticed the resemblance. That context matters. This wasn't an abstract policy tweak dreamed up in a planning meeting. It was a response to a visible, named mess on a storefront Anthropic doesn't even run, and it shipped the way most reactive fixes do: fast, narrow, and without much explanation attached.

No announcement post. No changelog line. Just a confirmation once someone asked.

Why Block This Now, Specifically?

Three things pushed this past "quiet guideline" into "enforced in the prompt." Steam's clone problem became visible enough that gaming press started writing about it by name, which turns a slow-burn concern into a reputational one almost overnight. Anthropic's own Usage Policy already gave it a ready-made justification, so there was no need to draft new policy, only to make Claude Code actually follow the policy that already existed. And Claude Code's reach into game development had grown enough that letting it recreate commercial titles stopped being a hypothetical liability and became a real one, sitting in front of anyone with a prompt box.

None of that is unique to Anthropic. Model providers have been tightening IP-adjacent behavior for a while now, usually quietly, usually after a visible incident rather than ahead of one. OpenAI and others have made similar narrow carve-outs around specific characters, specific franchises, specific named works, each time without much fanfare.

What's different here is how narrow the carve-out is. It doesn't ban games. It bans one specific behavior: recreating something that already exists and calling it new. That's a much easier line to defend in public than a blanket "no game development" rule would have been, and it's probably why Anthropic let Osmani confirm it on the record instead of staying silent on a reporter's question. A flat ban invites backlash from every solo developer using the tool legitimately. A narrow carve-out mostly just invites confusion about where the edge actually sits, and confusion is cheaper to absorb than backlash.

What Does the Rule Actually Block, and What Doesn't It Touch?

Here's where it gets messy, and where I'd be careful before assuming you're in the clear. The stated rule covers "a recreation of a specific existing game." By that wording, it doesn't cover games in a genre, your own platformer, your own match-three, your own roguelike, and it doesn't cover games built from your own assets and your own designs from the ground up.

What it doesn't clearly settle is longer: ports of a game you already shipped yourself, compatibility or interoperability work against an existing engine, decompilation done for research rather than resale, or mods that extend something you already legitimately own. Coverage on this split across the outlets that reported it. One account framed the policy as leaving room for inspecting software and doing compatibility work without triggering the block. IBTimes reported that Anthropic hadn't clarified how ports, decompiles, or mods get treated at all, and called the scope genuinely open.

Developer complaints back up that ambiguity in practice, not just in theory. Several people reported the refusal firing on commissioned, fully original projects that never touched an existing title's code or assets, just its genre conventions, a jump mechanic, a camera angle, a UI layout that happens to resemble something popular. That's the failure mode that actually matters if you're a solo builder. The model can't always tell "inspired by" from "recreation of," and when it guesses wrong, you get a flat refusal with no explanation and no setting you can flip to appeal it.

Anthropic hasn't said whether the rule applies only to Opus 5.5 or extends to other models you can run inside Claude Code, which makes it harder to simply route around by switching models for the session. Until that's clarified, I'd assume it travels with the product, not just the one model, and plan accordingly.

There's also no public appeal path. If a refusal fires on a project you're sure doesn't recreate anything, the options reported so far are limited to rewording the prompt or working around the model entirely for that one request. No ticket to open, no form to fill in, no confirmation that a human ever reviews the edge case. For a one-person studio shipping on a deadline, that's the part that actually costs time, not the policy's intent.

What Should a Solo Builder Actually Do About This?

If you ship games, or you're thinking about starting one, I'd treat this as a prompt-hygiene problem first, before I'd treat it as a legal one. Say explicitly, in your own prompt, what's original about the project: your own mechanics, your own art direction, your own naming from day one. Don't lean on "like [existing game] but different" as a shorthand for yourself, even casually, because that exact framing is what the new rule is tuned to catch.

If you're building a genre entry rather than something wholly novel, say the genre, not the game you're measuring yourself against. "A match-three puzzle game with a gravity twist" reads very differently to a prompt filter than "basically Candy Crush but mine," even when the two prompts describe almost the same build. Same intent, very different odds of getting through cleanly.

I don't build games with Claude Code myself. My own tools are smaller, utilities and workflow pieces, and none of them touch this rule at all. But if I did, I'd still log exactly where a refusal happened the first time I hit one. "No setting to disable it" doesn't mean there's no way to work around it. It means the workaround lives in how you phrase the request, not in a config flag you can flip once and forget. That's a worse developer experience than a toggle would be, and it's worth filing real feedback about if it catches legitimate original work, rather than just rewording the prompt and moving on silently.

I'd also keep a short written note, even a single line in a project README, describing what makes the game yours: the mechanic you invented, the art style you chose, the reason it isn't a reskin. Not because Anthropic reads your README. Because six months from now, when someone asks why your game looks a little like a genre classic, you'll want the answer already written down instead of reconstructed from memory under pressure.

For more on how Anthropic locks down Claude Code behavior at the prompt and config level rather than through a user-facing setting, the allowedProviders change in 2.1.285 and the working-directory write block added in 2.1.314 both follow the same shape: a narrow technical fix for one real incident, shipped quietly, with the explanation arriving only after someone asked for it.

Bottom Line

This isn't a ban on building games with Claude Code. It's a ban on one specific move: taking an existing game, swapping the name, and calling it original. That's a defensible line, and given what Steam has looked like these last few weeks, probably an overdue one too.

The part I'd actually worry about if I were shipping games right now is the enforcement gap, not the policy itself. A rule with no visible boundary and no way to disable it will catch legitimate original work alongside the clones it's aimed at, and developers are already reporting exactly that happening to them. If you're affected, the fix right now is prompt wording, not a setting buried in a config file. Say what's original about your project before you ask for the code, and keep the genre-versus-clone line explicit in your own head, because the model is being asked to draw that same line automatically, and it won't always get it right.

Opus 5.5 picked up this behavior as part of a broader run of changes since its launch, including the pricing and breaking changes covered here. So if you're still on an older Opus checkpoint inside Claude Code, don't assume this rule already applies to you. Don't assume it doesn't, either. Anthropic hasn't said, and until it does, the safest move is the same one either way: build the thing that's actually yours.