Last updated: November 3, 2026
Short answer: 5 steps fix a false Claude Code game refusal. First, check whether the refusal cites the game rule or a different safeguard. Then start a fresh session, reframe your request around your own mechanics and assets, and report the false positive with your version and model. The rule targets recreations of specific existing games, not original work.
Table of Contents
- What the Rule Actually Says
- How to Tell Which Refusal You Got
- Why a Working Session Suddenly Starts Refusing
- The 5-Step Fix
- What Still Works
- Is Claude Code Shutting Down?
- FAQs
What the Rule Actually Says
Anthropic added two sentences to Claude Code's prompt on Opus 5.5, pushed around October 27, 2026. The change was spotted by X user Chasm, who reported it arrived from Anthropic's servers under an internal flag name, and Anthropic's Addy Osmani confirmed it on October 30, 2026. There is no official Anthropic blog post about it, and there is no setting to switch it off.
The rule 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."
The stated reason is copyright. Osmani said using Claude to infringe intellectual property violates Anthropic's usage policy, and the company says original games, genre games, and games built from the user's own designs and assets are meant to keep working. Context, not excuse: Anthropic settled a major authors' copyright case for $1.5 billion in September 2025, so the company is guarding this line closely. Like other AI tools that changed overnight, covered in our Character.AI shutdown guide, a server-side change can alter your workflow without warning.
One honest warning before the fixes: this is a prompt-level steer, not a hard filter. That is why results are inconsistent, and why the same request can pass once and fail later.
How to Tell Which Refusal You Got
Not every Claude Code refusal comes from the game rule, and the fix depends on which one you got. Anthropic has not published official refusal wordings, so this table is built from reported cases, not from Anthropic documentation.
| What Claude Code said | What it usually means | Your next step |
|---|---|---|
| It cites recreating a game, or says a renamed game "isn't Tetris with a new name" | The game rule is the likely cause | Use the 5-step fix below |
| "Safeguards flagged this message" or "flagged this session" | A broader safeguard flag, reported most often on security-related work, not the game rule | Report it with /feedback, rewind to before the flagged turn, or start a new session; security researchers can apply for Anthropic's Cyber Verification Program |
| A vague refusal with no clear reason | Could be either cause | Test the same request in a fresh session with neutral wording before assuming the game rule hit you |
If your request names a specific existing game, the game rule is the first suspect. If it does not, and the work is reverse engineering, debugging, or security testing, the broader safeguard is equally likely.
Why a Working Session Suddenly Starts Refusing
This is the part other coverage skips, and it matters most. A developer who builds commissioned digital board-wargame modules for clients who own those games reported on Claude Code's GitHub tracker (issue #106167, October 30, 2026) that sessions which had done the same work for months suddenly began refusing it. The refusals began only after the developer asked Claude to check whether the new rule affected their work. From that point on, the refusal persisted in sessions working from the same project.
In plain terms: once the rule becomes part of a session's conversation, the session starts judging everything against it. Testing by AI Parabellum points the same way — short, clean test folders behaved better than long real projects. So a session that worked for months is not proof the rule will never touch it, and a refusal in one session is not proof every session will refuse.
Practical lesson: do not ask Claude, inside your working session, whether the game rule applies to your project. That question itself can poison the session.
The 5-Step Fix
Step 1: Diagnose the refusal
Read the refusal wording and match it to the table above. If it cites the game rule, continue here. If it is a safeguard flag, use the safeguard path instead: /feedback, rewind, or a new session.
Step 2: Start clean
Open a new session. If the refusal sits in one project, rewind to a checkpoint before the flagged turn, or move to a fresh folder. Do not carry the rule discussion into the new session — leave that conversation behind, or the new session inherits the same judgment.
Step 3: Reframe around ownership, honestly
Say the game is yours only if it truly is. Then stop naming the existing title and describe what you are building instead: the mechanics, the systems, the feel. "A turn-based tactics game with hex movement" passes the rule's own test far better than a named recreation.
Step 4: Prove it with your own assets
Put your design notes and art in the project and point Claude Code at them. Record lasting project details in your CLAUDE.md file. In AI Parabellum's tests, a game built from the user's own design file and art worked on every run, and a bug fix in the user's own game worked too.
Step 5: Report the false positive
If your own or licensed work is still refused, file a report on the Claude Code issue tracker with your version (check it with claude --version), your model, and a short prompt that shows the problem. Keep client details out of any public report. Osmani has also asked developers to send him false-positive examples directly. Note honestly: Anthropic has not published any ownership-verification or exemption program for game work yet, so a report is the documented path, not a guaranteed override.
What Still Works
Original games, genre games, and games from your own designs and assets are meant to keep working, and independent testing backs that up. AI Parabellum ran 18 tests across two Claude Code versions and found original games, generic platformers, own-design games, and bug fixes in the user's own game all worked. Named-game requests were not refused outright in their tests — they came back changed, with new names and art. The one clear pushback was a Tetris prompt on version 2.1.316, which answered that it was not Tetris with a new name and delivered a different puzzle game instead.
Separately, XDA Developers reported that Opus 5.5 refused to recreate Pokemon Diamond. So the honest picture is mixed: the rule is real, its effect is uneven, and a passing test does not make a recreation safe to release. Getting code from Claude Code gives you no rights to someone else's game.
Is Claude Code Shutting Down?
No. Claude Code is not shutting down, and Anthropic has announced no shutdown. This is a rule change, not a shutdown: the tool keeps working for original games, genre games, and games built from your own designs and assets. A refusal on one request is not the tool closing down.
The confusion is understandable, because the rule arrived overnight from Anthropic's servers, without warning. Tools do shut down with little notice — Google's Jules is the live example, going read-only on November 11, 2026 and shutting down fully on December 31, 2026, as covered in our Google Jules shutdown export guide. Google is also shipping new AI tools fast — our Gemini Omni 1.2 Flash guide shows how to make 15-second AI videos free in Google Vids. The lesson for every developer: keep your own code, design files, art, and prompts with you, never only inside one tool, so no rule change or shutdown can ever strand your project.
What Nobody Can Promise Yet
Three things remain unconfirmed, and any post claiming otherwise is guessing. First, Anthropic's statement covers Opus 5.5 only — nobody has published tests showing how other models behave under this rule. Second, the exact prompts that trigger a refusal are not documented, and the flag name reported by Chasm is unconfirmed by Anthropic. Third, your version alone does not settle it: the rule text appeared in version 2.1.316 but not 2.1.312, yet reports say it can also arrive from Anthropic's servers, so staying on an old version is not a fix and costs you other updates.
FAQs
What is the Claude Code game rule?
It is a two-sentence instruction added to Claude Code's prompt on Opus 5.5 around October 27, 2026. It tells Claude Code to build original games, genre games, and games from the user's own designs and assets, and not to build a recreation of a specific existing game, even if it is renamed or called original. Anthropic confirmed it on October 30, 2026, and there is no setting to turn it off.
Why did Claude Code refuse my own game?
Because the model judges surface signals, not ownership. If your request names an existing game or uses reverse-engineering terms, the model can treat it as a recreation attempt. Anthropic has not published any ownership-verification mechanism, so the model cannot check that the work is yours or licensed.
How do I know if the game rule hit me or it is another refusal?
If the refusal cites recreating a game or a renamed game, the game rule is the likely cause. If it says safeguards flagged the message or session, that is a broader safeguard flag with a different path: report it with /feedback, rewind, or start a new session. A vague refusal with no reason needs a fresh-session test before you blame the game rule.
My session worked for months, then started refusing. Why?
Most likely the session became poisoned. In a reported GitHub case, refusals began only after the developer asked Claude whether the new rule affected their work, and the refusal then persisted in that project. Once the rule enters a session's conversation, the session keeps judging work against it.
How do I recover a blocked session?
Start a new session, rewind to before the flagged turn, or move to a fresh folder, and do not discuss the rule inside the working session. Then reframe your request around your own mechanics and assets, with your design files in the project.
How do I reframe my request so legit work passes?
State ownership only if it is true, stop naming the existing title, describe the mechanics you want, and point Claude Code at your own design notes and art. Record lasting project details in your CLAUDE.md file so the project speaks for itself.
Can I turn the rule off or switch models?
No, there is no setting to turn the rule off, and it is delivered from Anthropic's servers as well as in newer versions. Anthropic's statement covers Opus 5.5 only, and no published test shows how other models behave, so switching models is not a confirmed workaround.
Where do I report a false positive?
File a report on the Claude Code issue tracker with your version, your model, and a short prompt that shows the problem, and keep client details out of public reports. Anthropic's Addy Osmani has also asked developers to send him false-positive examples. There is no published exemption program for game work yet.
Sources
- Anthropic confirmation via Addy Osmani's X post, as reported by Game Developer (October 30, 2026)
- Claude Code GitHub tracker, issue #106167: commissioned board-wargame modules refused (October 30, 2026)
- AI Parabellum: 18 test runs on Claude Code 2.1.312 and 2.1.316 (November 1, 2026)
- XDA Developers via ShiftDelete: Opus 5.5 refused a Pokemon Diamond recreation
- Gadget Review / VideoCardz / ProPakistani: reverse-engineering, modding, and debugging refusals
- Anthropic Usage Policy: ban on using its tools to infringe third-party intellectual property