Anthropic reportedly restricts Claude Code from cloning games, users report refusals

Anthropic has added guardrails to Claude Code that steer the tool away from recreating a specific existing game. The company confirmed the change in a post by employee Addy Osmani, who wrote that Anthropic "added two sentences to Claude Code's prompt on Opus 5.5" for that purpose. The reason given is that using Claude to infringe IP violates the company's usage policy. Tom's Hardware reported that the restriction is not limited to games. Users have reported refusals on reverse-engineering tasks as well, and an acquaintance of a Tom's Hardware staffer said a request to reverse engineer a product activation triggered safeguards when run through Claude Code on Opus 5.5.

Osmani's wording matters for what the rule actually covers. He said original games, genre games, and games built from a user's own designs and assets are meant to keep working, and that Anthropic would keep iterating to "strike the right balance." A screenshot posted by an X user named Chasm, which Tom's Hardware could not verify as the exact prompt text, shows a line telling the model not to build "a recreation of a specific existing game, even if it is renamed or called original." That phrasing targets a named product rather than a mechanic. It fits the pattern users describe: a request naming a franchise appears to trip the filter, while a request for a genre-level game with similar mechanics did not, according to one commenter on Reddit.

The reverse-engineering reports come from individual accounts. Tom's Hardware's Bruno Ferreira wrote that Jarred Walton, a former Tom's Hardware writer, said such work had run without problems until October 31 and that switching to Opus 4.8 let him continue. Tom's Hardware did not establish whether the reverse-engineering block is a separate rule or part of the same prompt change. It said the game-related blocking appears limited to Claude Code combined with Opus 5.5. A Redditor reportedly had an account suspended and then reinstated over using Opus 5.5 with a reverse-engineering tool, but the article gives no detail on that case.

Users describe a workaround that depends on the same limit. If the block depends on the Claude Code harness and the Opus 5.5 model, a user could in principle reach Opus through another harness or a wrapper, which is what Tom's Hardware suggested. A Reddit commenter said they were using Claude as a helper for DeepSeek to avoid the guardrails. The Tom's Hardware piece does not test this, and Osmani's post does not address workarounds. It is unclear whether Anthropic regards such routing as within the policy.

Commenters on the Tom's Hardware article raised a related objection about consistency. One argued that AI vendors that trained on copyrighted material are hypocritical to restrict cloning now, and another pointed out that competitors without the same limits could draw developers away. Those are opinions from readers, not findings in the reporting. The article itself does not assess whether the rule will hold up.