Users Say Claude Refuses to Clone Games or Reverse Engineer

User complaints are mounting that Anthropic's AI model Claude is refusing to help recreate existing games and with some reverse engineering work. According to claims from several users, Claude Code and/or the Opus 5.5 model are also blocking harmless reverse engineering attempts, including work done with the popular Reverse Engineer Anything (REA) tool.

It has been suggested that a written rule telling the model not to recreate a specific existing game may be behind the complaints. Anthropic has not officially confirmed that such a rule exists.

What the alleged rule says

According to text shared by the Twitter user Chasm, the rule works like this: original games, games belonging to a genre, and games built from the user's own designs and assets can be developed, but a copy of a specific existing game will not be produced, even if it is renamed or presented as original. Whether the text is the actual rule has not been verified, but it appears consistent with the clear refusals developers have run into over the past week.

The number of threads on Claude Code's GitHub page describing refusals on routine game-related jobs is also reportedly growing. The complaints include work being refused halfway through development, a game mechanic being replicated, and a board-game-style title being patched. Some of these reports may come from bad-faith users, but at least one is said to belong to a genuine developer.

A link to cybersecurity restrictions?

On the reverse engineering side, scattered user reports have also appeared on Reddit and other platforms. According to these, Claude refused to analyze a simple Windows crash dump, to remove bloatware from a Samsung TV, and to use REA, though these too remain user claims. Some game developers say that building anti-cheat systems for their own games, as well as security reviews, were blocked on the grounds that safety safeguards were triggered.

Anthropic's Cyber Verification Program, announced on October 6, restricts access to its frontier models' advanced cybersecurity capabilities to organizations within Project Glasswing, and it took effect immediately. The same rule set could also be responsible for the cybersecurity-related refusals. Still, it is notable that the refusal reports have only been arriving for the past week, even though 24 days have passed since the program launched.

Possible motives and industry reaction

Two possible reasons for the restrictions stand out. The first is that Anthropic and other AI companies may want to protect themselves in advance against lawsuits from software companies and game studios, since they offer tools that can reverse engineer software. There has been no official statement on this from Anthropic or the other companies. OpenAI is also reported to have recently disabled the account of a person who was trying to clean up the software on their own television. It is speculated that AI-assisted clones of application suites such as Adobe and Microsoft Office 365, along with countless recompilation projects for console games, worry software companies.

The second reason is abuse within the gaming world. A debate is under way over claims that Steam Next Fest has been flooded with AI-generated content copied from existing games. Some developers are also reportedly being extorted with threats that their game demos will be cloned into full games before the originals are finished and released, a practice said to target mainly small studios.

Online reaction was swift. Users are advising people who want to do reverse engineering work to use Chinese-origin, open-weight or locally run language models instead of Claude or GPT. The same comments point to the irony of frontier AI companies, which built their training data from the world's accumulated knowledge and art, now worrying about copyright infringement.