Claude Code’s New Restrictions Disrupt Game and Software Analysis Work

A coding agent that changes its boundaries can interrupt real work overnight. Developers say Claude Code, running Opus 5.5, has begun refusing requests involving existing games and software analysis. Anthropic has confirmed a prompt change intended to deter users from recreating specific games, but hasn’t explained whether that change accounts for the broader refusals.
The reports range from game development to reverse engineering. One developer says paid client work has stalled; other users report refusals during software analysis. At least one developer was able to resume a task by switching to an earlier Claude model.
Restrictions target game recreation
On Friday, Addy Osmani, who works on Anthropic’s technical team, said on X that Anthropic had changed Claude Code’s instructions for Opus 5.5.
“We 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 IP is against our usage policy.”
Osmani said the added guidance is meant to block recreations of specific games. Developers should still be able to make original games, work within established genres, and use their own designs and assets. He said Anthropic would keep adjusting the instructions and invited developers to contact him if Claude refuses work based on their own creations.
That distinction may be straightforward in principle and difficult to apply in practice. Some commissioned projects involve games that already exist, even when developers use original code and materials supplied by the client. Anthropic hasn’t publicly said whether reports of refusals on reverse-engineering tasks stem from the game instructions or from separate cybersecurity safeguards.
Software analysis is caught in the uncertainty
Jarred Walton, a technical writer at Phison Electronics USA, ran into refusals while testing software he was developing, according to his account of the issue. Switching to Opus 4.8 let him continue. The change in behavior leaves users with little visibility into why a request fails or which instruction prompted the refusal.
The consequences are concrete for Ray Weiss, founder and CEO of Conflict Simulations LLC. On October 30, he reported on GitHub that Claude Code had stopped supporting paid client projects it had helped with for months.
Weiss’s company builds digital versions of board wargames for clients who own the physical games. He says the work uses original code and client-provided materials, yet Claude Code now declines to continue because the projects involve recreating existing games. Weiss asked Anthropic to distinguish commissioned work from infringement, establish a way to verify legitimate projects, and document significant changes to the tool’s instructions.
Hosted agents can change underfoot
The wider issue is control. With a hosted coding agent, a provider can change what the system accepts without any change to a developer’s code, tools, or prompts. A workflow that worked yesterday may hit a refusal today, and users may have to switch models or rework their process before they understand why.
Anthropic’s cybersecurity policies have also had to account for legitimate work. On October 6, the company expanded its Cyber Verification Program, giving vetted security researchers access to fewer restrictions after acknowledging that safeguards can block valid cybersecurity tasks. It’s not clear whether those safeguards are involved in the current complaints. Developers’ reports that older Claude models handle some of the same tasks add to the uncertainty.
Reverse engineering has practical uses
Reverse engineering isn’t limited to security testing. In October, Red Hat engineering director Christian Schaller used Claude Code to help reverse engineer a scanner driver, adding Linux support for hardware whose manufacturer had changed its chipset.
Anthropic has acknowledged that the game-recreation rules need refinement. But developers still lack a clear account of why some software-analysis requests now fail, whether the refusals share a cause, and what other established workflows might be affected. The practical test for Anthropic will be whether it can explain those boundaries clearly enough for developers to rely on Claude Code in paid and technical work.
The New Stack contacted Anthropic to ask whether the refusals relate to the game restrictions or other safeguards. The company did not respond before publication.
The post Developers are downgrading their AI on purpose — and it’s working appeared first on The New Stack.