What Anthropic confirmed
X user Chasm said they spotted an update to Claude Code's "rules" that was pushed around October 27. The rule sits under the identifier "tengu_heron_brook". It tells the model to build original games, genre games and games from the user's own designs and assets. It also tells the model not to recreate a specific existing game, even if renamed or called original. Chasm reports there is no way to disable it.
Tom's Hardware first described this rule as unconfirmed. That is no longer the case. An Anthropic spokesperson pointed Game Developer to a statement by Anthropic technical staff member Addy Osmani. He wrote that two sentences were added 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 the usage policy. He added that original games, genre games and games built from your own designs and assets are meant to keep working. He also said Anthropic will keep iterating to get the balance right.
Here is what the confirmation does not cover:
- The confirmed text describes a prompt change, not a published policy page about games.
- Anthropic has not said how it defines a "recreation".
- Anthropic has not said how Claude would judge whether a user owns a game idea or its assets. Reporting notes that Anthropic did not explain how Claude will determine whether a user owns the game, concept or assets they ask it to use.
- Game Developer found no published exemption process for game developers.
How far the rule reaches
The scope is still murky. XDA's testing suggests two delivery paths. Opus 5.5 can receive the game-recreation instruction independently, while accounts that get it through tengu_heron_brook have it applied to all models used in Claude Code. XDA says it is unclear how widely Anthropic has deployed the instruction, whether it will become permanent across all models, or which kinds of game development count as a recreation.
XDA also tested Opus 5.5 on a decompilation question. Asked about a Pokémon Diamond decompilation project, the model agreed to help and explicitly distinguished reverse engineering from recreating a game. That is one outlet's test of one request. It does not show how every request will be handled. It does suggest the rule is aimed at building clones rather than at analysing code in general.
A concrete false-positive complaint
A GitHub issue opened on October 30 shows the likely friction. The reporter says they run Conflict Simulations LLC. They say they build personalised digital board-wargame modules for individual clients who own the physical games.
- They say the code is original and written from scratch.
- They say clients supply the materials and prove ownership first.
- They say the builds are for the client's private use and never distributed.
- They say sessions that did this work for months now refuse to continue.
- They corrected an earlier claim: the work is not publisher-licensed.
- They say they honour publisher and designer opt-outs.
- They say they won't build modules for games publishers have already digitised.
Treat all of this as the reporter's own account. It has not been independently verified, and the retrieved issue page showed no Anthropic response.
The reporter's timeline is odd. They say the work kept running after the rule reportedly shipped. They say refusals began only after they asked Claude whether the rule applied to their work, and persisted in sessions using the same repositories. One commenter suggested the cause was memory or Git history carrying the refusal forward. That is a guess, not a finding.
The reporter also says Anthropic's support bot claimed the policy applies to all customers without exception and that no verification path exists. They say their tickets were later passed to humans. This is secondhand and not a company statement.
Their requests were:
- Narrow the rule to actual infringement.
- Add a verification path for professionals.
- Document server-delivered instruction changes in the Claude Code changelog.
The last request raises a real operational point. Claude Code behaviour can change through server-delivered instructions, so no client update or changelog entry is needed. Teams that depend on it mid-project can be caught out.
Why Anthropic did it
Game Developer reports that Anthropic appears to be responding to the flood of AI-driven game remakes in recent months. Digital Sun Games was among several studios that cancelled a public playtest of its upcoming game ReVamp, fearing the demo could be fed into an AI tool. That is context, not evidence that a ReVamp clone exists.
The same outlet doubts the rule will stop determined cloners. It says the rule may deter only the most low-effort copycats, and notes that safety limits on generative AI have repeatedly been sidestepped with rhetorical tricks. It also notes that users can switch to open-weight models without such restrictions.
Keep the security claims separate
Tom's Hardware ties the game rule to reported refusals on a Windows crash dump analysis, Samsung TV cleanup, use of the Reverse Engineer Anything tool, and anti-cheat and security reviews. The evidence does not show the game rule causes any of these. They are user claims from Reddit and elsewhere.
There is a separate, documented safeguard. Anthropic's Opus 5.5 announcement says it deploys the model with cybersecurity safeguards similar to those on Fable 5.1. Routine bug finding and fixing is allowed, but most cybersecurity tasks are re-routed to Opus 4.8. Verified practitioners can apply through the Cyber Verification Program. That program is a security access route, and nothing indicates it offers an exception for game development.
What developers and admins should do
- Record the refusal. Capture the exact wording and the model in use. Osmani invited developers to send him cases where Claude turns down something that is theirs.
- Read the scope literally. Original games, genre work and projects built from your own assets are meant to be unaffected. Faithful ports, remakes or rebuilds of a named title are the target. Renaming won't help.
- Check your model and version. The rule reportedly appears in Claude Code v2.1.314. Opus 5.5 is also available on Microsoft Azure, so teams using it through Azure should expect similar prompt-level behaviour. That is an inference. Anthropic has not said how the rule applies on other platforms.
- Plan for server-side changes. If AI tooling is in your pipeline, behaviour can shift without a client update.
Bottom line
The confirmed fact is narrow: a two-sentence prompt addition to Claude Code on Opus 5.5 that discourages recreating specific existing games. The messier claims are not confirmed. These are the reports of blocked reverse engineering, crash dump analysis and anti-cheat work, and the one board-game developer's complaint. Anthropic has promised to iterate, so the boundaries are likely to shift. Until it explains how it tells a clone from legitimate work, expect some false positives.
Update: Developer reports Opus 5.5 refusal, switches to Opus 4.8 (November 2, 2026)
According to Tom’s Hardware, developer Jarred Walton said Claude Code stopped helping him run checks on software he is developing after previously performing the work without issue through October 31. Walton told the outlet that switching from Opus 5.5 to Opus 4.8 let him continue.
Tom’s Hardware also reported receiving screenshots from a staff member’s acquaintance showing Opus 5.5 in Claude Code triggering safeguards when asked to reverse engineer a product activation. The report does not establish whether that refusal and Walton’s experience share a cause.
Walton speculated that Anthropic may have added a separate reverse-engineering rule, but Tom’s Hardware did not confirm one. The reports also do not establish that the game-recreation instruction caused either refusal.
For developers and IT teams, Walton’s account suggests that changing models may help with a particular blocked workflow, but it does not show that Opus 4.8 will accept every such request. Teams encountering refusals should record the model, request and response before testing an alternative.
Update: Claude Code revises game-recreation rule (November 5, 2026)
Anthropic technical staff member Addy Osmani says the game-recreation rule has been updated in Claude Code 2.1.317. The change is intended to let developers name an existing game when describing the mechanics they want, while still producing an original game.
Osmani acknowledged that the earlier rule was blocking some projects it should have allowed, including requests that referenced a game only to explain desired mechanics. Under the revised rule, Claude should build an original game in the requested genre rather than reject such a reference outright.
Requests to copy a specific game remain restricted. Osmani says Claude will now offer to build something similar instead.
For developers, the distinction is practical: naming a game as a design reference should no longer automatically prevent Claude Code from helping. The announcement does not establish how reliably the revised rule will distinguish a legitimate reference from a copy, or whether it resolves the board-game developer’s reported refusal.
References
- Anthropic's Claude AI blocks video game cloning and reverse-engineering Tom's Hardware · 2026-10-31T12:20:00+00:00
- Anthropic confirms new Claude Code limits to prevent game cloning — users report Opus 5.5 is also actively blocking reverse engineering tasks Tom's Hardware · 2026-11-02T12:30:00+00:00
- Server-delivered "games" instruction blocks professional development work #106167 github.com