Anthropic Confirms Claude Code Guardrails Block Game Cloning

Anthropic has confirmed it added two sentences to Claude Code's system prompt on the Opus 5.5 model to block the recreation of existing video games, and users now report the same guardrails are refusing reverse engineering work as well.

The confirmation came directly from Addy Osmani, an Anthropic employee, who responded publicly to a tweet containing the purported rule update. "[Anthropic] 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 [its] usage policy," Osmani wrote. He added that work on original games is intended to keep functioning, and that the company is iterating on the rules to try to "strike the right balance."

The exact wording of those sentences remains unpublished. However, Osmani's phrasing suggests they match a screenshot posted by X user Chasm, which reads: "Do build original games, games in a genre, and games from the user's own designs and assets. Don't build a recreation of a specific existing game, even if it is renamed or called original."

What exactly is being blocked?

The restrictions appear narrow in scope, at least for now. Evidence gathered so far indicates the game-related blocking is limited to the combination of Claude Code, Anthropic's command-line coding harness, and the Opus 5.5 model. Users running Opus through alternative harnesses such as Ollama, or routing requests through a wrapper, appear able to sidestep the limitation. The rule is prompt-level steering, not a model-wide refusal.

The reverse engineering blocking is less clearly bounded. Complaints had already surfaced on Claude Code's own GitHub repository, where users reported the assistant refusing to work on their games and on reverse-engineering tasks. One direct confirmation came from Jarred Walton, a developer and former contributor to the original publication, who was running software checks of his own product. He states the work proceeded without issue until October 31, and that switching to Opus 4.8 allowed him to continue.

Walton theorizes Anthropic added a new rule to Code specifically covering reverse engineering. The timing supports the theory. The wave of refusal reports all appeared within roughly the same week, suggesting a coordinated policy change rather than isolated model quirks.

How far do the safeguards reach?

Further reports continue to accumulate. A staffer's direct acquaintance sent screenshots showing a request to Opus 5.5, made through Code, to reverse engineer a product activation; the request triggered safeguards. On Reddit, a user claims an account suspension over using Opus 5.5 with Reverse Engineer Anything, an automation tool. The account was reportedly reinstated afterward.

The suspensions fit an existing structure. Anthropic operates a Cyber Verification Program, under which only approved entities receive access to the full set of frontier models' cybersecurity capabilities. If the new refusals tie into that program, ordinary developers doing legitimate compatibility and verification work may find themselves filtered by rules intended for a narrower threat model.

The distinction matters commercially. A developer auditing their own software, as Walton was, is not attempting an IP infringement. Broad prompt-level refusals that cannot distinguish the two cases impose a real cost on paying subscribers, and the community reaction has been sharp accordingly.

What happens next?

Anthropic says it is still iterating on the wording, so the blocking behavior may soften as the company refines the balance between IP protection and legitimate use. For now, developers on Opus 5.5 face refusals on two fronts, and the reliable workaround is a downgrade to Opus 4.8 or a different harness entirely.