Skip to content
, ,

Claude Code Won’t Clone Existing Games Now: The New Opus 5.5 Rule Explained

8–12 minutes
Shocked woman in orange hoodie holding card with Claude logo in a server room, text No game clones Claude Code

If Claude Code suddenly said no to your game project this week, you are not imagining it. Anthropic has added a new rule that stops Claude Code from recreating someone else’s specific video game, even if you rename it.

The change follows a viral wave of AI-built knockoffs, including browser ports of famous shooters and open-world games. Here is what the rule says, what you can still build, why some legitimate developers are getting caught, and what to do if Claude refuses your own project.

TL;DR

  • What changed: Anthropic added “two sentences” to Claude Code’s system prompt for Opus 5.5 to steer it away from cloning existing games, confirmed by Anthropic’s Addy Osmani on 30 October 2026.
  • Still allowed: original games, games in a genre (a platformer, a roguelike) and games from your own designs and assets.
  • Blocked: a recreation of a specific existing game, even when it is renamed or called “original”.
  • Why: users built hundreds of unofficial clones with Opus 5.5 and GPT-6 Astra, and Anthropic says IP infringement breaks its usage policy.
  • Side issue: some users also report reverse-engineering refusals. Anthropic has not confirmed a new rule for that; it may be the existing cyber safeguards.
  • Wrongly refused? Run /feedback in Claude Code and keep your own design docs in the project.

What Happened

On 30 October 2026, Anthropic employee Addy Osmani posted on X that the team had “added two sentences to Claude Code’s prompt on Opus 5.5 to steer it away from recreating someone else’s specific game.”

He said using Claude to infringe intellectual property already breaks Anthropic’s usage policy. He also said the team will keep iterating to “strike the right balance”, and invited developers to send examples where Claude refused work on something they own.

Tom’s Hardware reported the change on 2 November, and The Decoder followed on 3 November. The story spread fast among indie developers and on Reddit.

Quick answer: Claude Code will still help you make games. It will not help you rebuild a specific game that someone else owns.

The AI Knockoff Wave Behind It

Frontier coding agents got good enough this autumn to rebuild large chunks of famous games from memory, screenshots and decompiled code. People noticed, and shared the results.

  • Browser game ports: users reportedly made hundreds of unofficial browser versions of well-known titles, including Halo and GTA Vice City, with Opus 5.5 and GPT-6 Astra (The Decoder).
  • Reverse Engineer Anything: an open-source tool that pairs AI agents with reverse engineering to turn native binaries into readable pseudocode.
  • App clones: one developer recreated parts of Adobe’s Creative Cloud apps.
  • Emulators: after a PlayStation 5 jailbreak, a quick wave of AI-assisted emulators followed.

Big game publishers are known for protecting their IP, and an AI company whose tool is producing copies of their games has an obvious legal risk. The rule is Anthropic’s response.

What the Rule Actually Says

Anthropic has not published the exact wording. What we have is Osmani’s description plus a screenshot that X user “Chasm” shared, which Tom’s Hardware reports may be the two sentences.

The circulating wording, which is not confirmed by Anthropic, reads:

  • “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.”

Note: Game Developer also reported a feature flag name and a rollout around 27 October. Neither has been confirmed by Anthropic, so treat them as unverified.

Where the rule applies

Surface Game rule? Source
Claude Code + Opus 5.5 Yes (confirmed) Osmani on X, Tom’s Hardware
Claude Code + older models (Opus 4.8) Not reported Tom’s Hardware
Claude.ai chat Not confirmed either way SSD Nodes summary
Claude API Not confirmed; the usage policy still bans IP infringement Anthropic usage policy

Allowed vs Blocked: Examples

Based on Osmani’s description, here is how typical requests should land. These are my examples, not Anthropic’s official list.

Request Expected result Why
“Build a 2D platformer with wall jumps and my own pixel art” Allowed Genre game, your assets
“Make a roguelike deck-builder with a space theme” Allowed Genre game
“Turn my Figma game design doc into a playable prototype” Allowed Your own design
“Add multiplayer to my existing Godot project” Allowed Your code
“Recreate [famous game] level by level in the browser” Blocked Specific existing game
“Clone [famous game] but call it something else” Blocked Renaming does not change it
“Rebuild the original [famous game] menus, characters and music” Blocked Copies protected assets

Tip: Describe the mechanics you want, not another game’s name. “A top-down racer with drift boosts” gets a better and more original result than “make it like [famous game]”.

Timeline

Date (2026) Event
6 Oct Anthropic announces its Cyber Verification Program for security work
~27 Oct Game rule reportedly rolls out (unconfirmed)
30 Oct Osmani confirms the two-sentence change on X; Game Developer reports it
31 Oct Some users report reverse-engineering checks blocked from this date
2 Nov Tom’s Hardware reports the rule and the refusals
3 Nov The Decoder reports the knockoff wave context
12 Nov Anthropic’s updated usage policy takes effect

The Reverse-Engineering Refusals

Alongside the game rule, developers have reported Opus 5.5 refusing some reverse-engineering work in Claude Code. This part is less clear.

  • Software checks: former Tom’s Hardware writer Jarred Walton said his reverse-engineering checks worked until 31 October, then stopped on Opus 5.5 but worked on Opus 4.8.
  • Product activation: one user reported a refusal on a request to reverse engineer a product activation.
  • Account suspension: a Redditor reportedly had an account suspended, then reinstated, after using Opus 5.5 with Reverse Engineer Anything.

Anthropic has not confirmed a new reverse-engineering rule. Reports point to Claude Code’s existing cyber safeguards, which can flag tasks like exploit generation or binary vulnerability scanning and move the session to another model.

Warning: Do not try to trick the model around a refusal. Anthropic’s usage policy applies on every surface, and repeated attempts can get an account flagged or suspended.

What a Refusal Looks Like

There is no confirmed sample of a game-rule refusal yet. Based on reports, you may see one of these.

  • A plain text answer: Claude explains it can help with an original game instead.
  • A usage policy error: API Error: <model> can't help with this. Start a new session to continue. with a Request ID.
  • A cyber flag: <model>'s safeguards flagged this session, or a note that the session switched to another model.
  • Over the API: a response with stop_reason: "refusal".

How to Keep Building Your Own Game

Most refusals for honest projects come from vague prompts that name a famous game. These steps help Claude see your project for what it is.

  1. Put your design doc in the repo. Add a GAME_DESIGN.md with your own title, story, characters and mechanics, and mention it in your CLAUDE.md.
  2. Keep your own assets in the project. Sprites, sounds and level files you made show the work is original.
  3. Describe mechanics, not titles. Say “physics-based slingshot puzzle” rather than naming the game that made it popular.
  4. Use plan mode first. Ask Claude to plan the feature before coding with plan mode, so you can fix any misunderstanding early.
  5. Roll back a flagged turn. Press Esc twice or run /rewind to go back to a checkpoint, then rephrase. Resending the same message will not help.
  6. Report false positives. Run /feedback in Claude Code with a short description. Osmani asked for exactly these reports.

For security researchers, the official route is Anthropic’s Cyber Verification Program at portal.anthropic.com/programs. Anthropic aims to decide within seven business days, but its pages disagree on whether Opus 5.5 is covered yet.

Tip: New to Claude Code? Start with my install guide and the commands cheat sheet.

Do Other AI Coding Tools Have the Same Rule?

Only Anthropic has announced a game-specific rule. But every major provider bans IP infringement in its terms, so “the other tool allowed it” does not make a clone legal.

Tool Game-clone rule announced? Notes
Claude Code (Opus 5.5) Yes Two-sentence system prompt rule
OpenAI Codex (GPT-6 Astra) No public rule found OpenAI’s usage policies ban IP infringement
Cursor No Inherits the policies of the model you choose
OpenCode No One forum user says it has no such issue; it still uses providers’ models and terms (see OpenCode vs Claude Code)
Local models (Ollama, LM Studio) No No provider filter, but copyright law still applies to what you release

Game mechanics on their own are generally hard to protect. What is protected is the expression: art, characters, names, music, story, level layouts and code.

  • Fine in general: a new game with your own art that shares a genre or mechanic.
  • Risky: copying characters, logos, maps, sounds or recognisable level design.
  • Trademark trouble: using a famous name or look-alike logo, even in a free browser game.

This is a general explainer, not legal advice. If you plan to sell a game that is inspired by another, talk to an IP lawyer.

What Is Still Unclear

  • The exact wording of the two sentences.
  • Whether Claude.ai and the API apply the same game rule.
  • Whether the reverse-engineering refusals are a new rule or the existing cyber classifier.
  • How often original projects are wrongly blocked. Users have filed complaints on Claude Code’s GitHub, but there is no official count.

My Take

I think this rule was inevitable. When an AI agent can rebuild a famous game in a weekend and people post the results publicly, the company behind the agent becomes the obvious legal target.

The rule itself is narrow and sensible: make your own game, not someone else’s. The real cost is false positives, so the faster Anthropic tunes it using /feedback reports, the better.

For indie devs, my advice is simple. Write your design doc first, keep your own assets in the repo, and describe mechanics rather than titles. You will rarely hit the rule.

FAQ

Can Claude Code still make games?

Yes. It can build original games, genre games and games from your own designs and assets. It only declines recreations of specific existing games.

Which model has the game-clone rule?

Anthropic confirmed it for Claude Code with Opus 5.5. Reports do not show it on older models such as Opus 4.8.

Does renaming a clone get around the rule?

No. The reported wording says a recreation is blocked even if it is renamed or called original.

Is Claude also blocking reverse engineering?

Some users report refusals, but Anthropic has not confirmed a new rule. The likely cause is Claude Code’s existing cyber safeguards.

Claude refused my own original game. What should I do?

Add your design doc and assets to the project, describe mechanics instead of other games’ names, use /rewind to rephrase, and send a /feedback report.

Can my account get banned for asking for a clone?

A single refused request is unlikely to cause problems, but IP infringement breaks Anthropic’s usage policy. Repeatedly trying to get around refusals is what risks action.

Is it legal to make a game “inspired by” another game?

Usually, if the art, characters, story, names and code are your own. Copying protected assets or trademarks is where legal trouble starts.

Sources

Free weekly email

Get new AI guides every week

One short weekly email with our newest local AI, ComfyUI and AI tool guides. No spam, no pop-ups, just the good stuff.

Free. Unsubscribe anytime in one click. We only use your email to send this newsletter. See our privacy policy.

Shocked woman in orange hoodie holding card with Claude logo in a server room, text No game clones Claude Code