Anthropic blocks Claude Code from recreating existing video games

Anthropic has confirmed that it changed Claude Code so the tool stops rebuilding other people’s video games, adding a rule to the coding tool’s prompt on Opus 5.5 after weeks of AI-assisted clones unsettled indie studios. Original games, genre games and games built from a user’s own designs and assets are meant to keep working, but developers have filed GitHub issues reporting refusals during ordinary game work, and some users report separate refusals on reverse-engineering and debugging tasks.

What Anthropic changed in Claude Code for Opus 5.5

Anthropic did not publish a standalone announcement. According to Tbreak Media, which cites reporting by Game Developer, an Anthropic spokesperson instead pointed to a post on X by Addy Osmani, an Anthropic member of technical staff. Osmani said the company 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 intellectual property (IP) is against Anthropic’s usage policy. He added that Anthropic will keep iterating to get the balance right.

A system prompt is the standing set of instructions a tool sends to the model with every request, so a change there alters behaviour without retraining the model. Tbreak Media notes that Opus 5.5 is the model Anthropic launched with lower prices and stronger safeguards, and that the company’s usage policy has been tightened more than once this year.

Detail	What the sources say
Tool and model	Claude Code, prompt on Opus 5.5
Size of change	Two sentences, per Osmani
Where it surfaced	Rules entry listed under “tengu_heron_brook”, pushed on 27 October
Stated reason	Using Claude to infringe IP is against the usage policy
User toggle	None, according to the X user Chasm
Other models in Claude Code	Not addressed by Anthropic

What the rule says, and what is still unconfirmed

The change surfaced when an X user going by Chasm spotted a new entry in Claude Code’s rules, listed under “tengu_heron_brook” and pushed on 27 October. As Chasm reported it, the rule tells the tool to build original games, games in a genre, and games from the user’s own designs and assets. Tbreak Media renders the restrictive half as:

“Do not build a recreation of a specific existing game, even if it is renamed or called original.”

Tom’s Hardware quotes the same line slightly differently, beginning “Don’t build a recreation of a specific existing game”. More importantly, Tom’s Hardware says there is no official or additional confirmation that this is the actual rule text, though it calls it plausible given the clear refusals developers have encountered over the past week. In short, Anthropic has confirmed that a game-cloning instruction exists and why; the exact wording circulating online comes from Chasm’s report and has not been confirmed by the company.

The clause about renaming is aimed at the most common workaround. Tbreak Media points out that a rename, the usual first dodge for cloners, is covered explicitly.

Why games, and why now

The rule follows a difficult few weeks for independent developers. Tom’s Hardware reports that some game developers have recently been hit by extortion scams threatening to clone their game demos into complete games before the originals can be finished and released. Tbreak Media reports that Digital Sun Games cancelled the public playtest for its upcoming game ReVamp in case someone fed the demo into an AI tool and shipped a copy first, and that, according to This Week In Videogames, Drunken Goddess Reflux did the same; both studios opted for smaller private tests instead.

Developers have also pushed Valve to say how it will stop clones from displacing their games on Steam. At the Steam Indie Connect event in Busan, South Korea, on 27 October, Valve’s Tom Giardino and Marc Wilkinson acknowledged the problem, but according to PC Gamer’s reporting cited by Tbreak Media, Valve does not want to be a gatekeeper on which tools developers use and announced no new rules. Tbreak Media describes Anthropic’s change as the first move on the issue from an AI vendor itself. Developers who receive threats to clone their games unless they pay are being asked to report them to Valve.

What developers say the restriction is blocking

Tom’s Hardware says the existence of a game-related rule is real, pointing to an influx of issues on Claude Code’s GitHub repository complaining about refusals to do standard game work halfway through development, when replicating a game mechanic, and when patching a tabletop-like game. The outlet acknowledges that some of those reports may come from users with nefarious purposes, but says at least one is from a bona fide developer with a real track record.

A second, overlapping set of complaints concerns reverse engineering, the practice of analysing compiled software to understand how it works. Tom’s Hardware reports user claims that Claude Code and/or Opus 5.5 are blocking innocuous reverse-engineering attempts, including those using the popular Reverse Engineer Anything (REA) tool. Reports on Reddit and other platforms allegedly describe Claude refusing to analyse a simple Windows crash dump, to clear junkware from a Samsung TV, or to use REA at all. These reports do not all involve games, and they remain user claims.

Some purported game developers also say Claude blocked anti-cheat development and security reviews for their own games, with the model citing triggered safeguards. Tom’s Hardware notes that the timing of the game and cybersecurity restrictions may be coincidence. Anthropic introduced its Cyber Verification Program on 6 October, under which access to its frontier models’ advanced cybersecurity capabilities would be limited to covered entities. The sources do not establish which refusals come from the game rule and which from the cybersecurity safeguards.

What it means for builders using Claude Code on game projects

For most original game work, Anthropic’s stated intent is that nothing changes: original titles, games in an established genre, and games built from a developer’s own designs and assets should still be supported. Tbreak Media says that a refusal on your own game is the case Anthropic wants to hear about.

The grey zone is work that touches someone else’s game. Claude Code has been used for exactly that kind of project. In one example described by MindStudio on 10 October, a developer built a Red Dead Redemption 2 tornado mod almost entirely through Claude Code with Opus 4.5, with the model handling particle effects, NPC reactions and a configurable mod menu over roughly 10 to 15 rounds of testing and feedback. That mod adds a system the base game lacks rather than recreating the game. In another, a public GitHub project documents reverse engineering a 1980s Commodore 64 strategy game, Weltendämmerung by Dirk Maier, and porting it to the web with Claude Code and Opus 4.5. Its author says the work took three days, with a total API cost of $100 across more than 50 sessions.

Both examples used Opus 4.5, and none of the sources tests how the Opus 5.5 rule treats mods, ports or preservation work of this kind. Builders whose workflows resemble them, especially anything that reproduces an existing game’s mechanics or structure, should expect the possibility of refusals and keep evidence of ownership or permission close at hand, since it is unclear how Claude would judge whether a user owns a game idea or its assets.

Limits and open questions

Game Developer’s own read, as summarised by Tbreak Media, is that the rule may only deter the lowest-effort copycats. Researchers and reporters have repeatedly shown that safety limits on generative AI tools can be sidestepped with rhetorical tricks. Users replying to Chasm also noted that anyone who finds Claude too restrictive can switch to open-weight models with no such limits, which means the rule shifts copycat activity rather than ending it.

The open questions the sources raise:

• Scope: Anthropic has not said whether models other than Opus 5.5 in Claude Code are affected.
• Control: Chasm reports there is no way to switch the rule off.
• Edge cases: a fan project, a homage or a clone-in-all-but-name could sit on either side of the line.
• Wording: the exact text of the rule has not been officially confirmed.
• Attribution of refusals: whether reverse-engineering and anti-cheat refusals stem from the game rule or from cybersecurity safeguards remains unclear.

What to watch next

Osmani’s commitment to keep iterating suggests the rule will change; his reference to striking the right balance indicates Anthropic expects to adjust it. Signals to track: whether Anthropic publishes the rule text or documents it officially, whether it clarifies which models it covers, how it responds to the GitHub issues from developers reporting refusals on their own projects, and whether it separates game-cloning limits from the cybersecurity restrictions behind its Cyber Verification Program. On the platform side, watch whether Valve moves beyond acknowledging the clone problem, and whether other AI vendors adopt similar limits. For related context on how Anthropic is adding guardrails at the model level, see Anthropic trains Claude to flag possible misuse after each reply.