Image: Tom's Hardware
Why it mattersA paid Claude Code account can lose a feature it had yesterday. There is no changelog entry and no appeal path.
Developers paying for Claude Code began getting refusals on their own game projects this week. There was no changelog entry warning them. Tom's Hardware reports that Claude Code and the Opus 5.5 model are blocking three kinds of work. The three are video game recreation, reverse-engineering, and some everyday troubleshooting. The outlet says the trigger is a new server-side rule. Anthropic has not confirmed it.
The paper cites a GitHub issue filed by a developer who runs Conflict Simulations LLC. He writes that his business builds digital versions of board wargames under paid commission. The clients already own the games. Sessions that had done this work for months now refuse to continue it. He says the model cites the new rule against reproducing games.
What the rule may say
Tom's Hardware prints what X user Chasm says the model's prompt now contains. The quoted text has two parts. First part: "Do build original games, games in a genre, and games from the user's own designs and assets". Second part: "Don't build a recreation of a specific existing game, even if it is renamed or called original". The outlet says there is no official confirmation of this text. It adds that the behaviour on the Claude Code issue tracker matches it.
The paper lists five open issues on the anthropics/claude-code repository. Developers report mid-project refusals in each one. One is a plain game work refusal. One is a game mechanic replication refusal. One is a tabletop-game patch refusal. One is a security review blocked on the developer's own anti-cheat code. One of the reporters names Opus 5.5. He says the problem spans two paid accounts on multiple machines.
The reverse-engineering side
Tom's Hardware says a second strand of refusals covers reverse-engineering. It includes refusals with the popular Reverse Engineer Anything tool. The paper lists Reddit reports of Claude declining a Windows crash dump analysis. It also lists refusals on a Samsung TV junkware cleanup and generic Reverse Engineer Anything prompts. The outlet notes that Anthropic recently published a Cyber Verification Program. That program routes advanced cybersecurity work through Project Glasswing. The paper suggests the same ruleset may be causing both patterns.
Anthropic has made no public statement on the new behaviour. Tom's Hardware also notes OpenAI recently deactivated the account of a user working on their TV's software.
What a team on Claude Code should do today
The shape of the problem is a server-side rule that can change at any time. It can change what the model refuses with no changelog entry. There is no appeal path. The GitHub issue author asks for a verification or exemption path. He wants one for commissioned work with proof of ownership, like the one for vetted security researchers.
The issue author also asks Anthropic to narrow the rule. He wants it to target actual infringement only, meaning reproducing copyrighted code, assets or text. He also asks Anthropic to document server-delivered instruction changes in the Claude Code changelog. Until that happens, a team that depends on Claude Code for game work or routine reverse-engineering has one option. Pick a different tool for the affected tasks. Keep a working alternative ready for the next silent change.
Source
Tom's Hardware report by Bruno Ferreira, with evidence in Claude Code GitHub issue #106167 and the four related open issues it links.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
