Can Anthropic ban AI slop games?

Independent studios have started cancelling public playtests. Digital Sun Games is one of several to pull one, on the reasoning that showing a game to the public now invites a fast-follow AI remake that beats the original to market. That is the thing most coverage of Anthropic's new games rule has left out, and it is the reason the rule exists.

On 27 October, Claude Code picked up a server-pushed instruction under a flag called tengu_heron_brook, spotted by the X user Chasm: build original games, games in a genre, and games from the user's own designs and assets, but do not build a recreation of a specific existing game, even if it is renamed or called original. There is no way to switch it off.

Addy Osmani, a member of Anthropic's technical staff, confirmed it. Two sentences added to Claude Code's prompt on Opus 5.5, because using Claude to infringe IP breaches the usage policy. Original, genre and own-asset games are meant to keep working, the company will keep iterating on the balance, and anyone whose own project gets refused should send it to him. An Anthropic spokesperson referred Game Developer to that statement rather than issuing another.

So the motivation is sound. Over recent months users have been rebuilding both classic and unreleased games with AI, for amusement and for money, and PC Gamer frames the change as following weeks of Steam being inundated with AI slop. The developers most exposed are precisely the small studios who would otherwise be loudest about a model provider restricting their tools. A rule aimed at fast-follow cloning is defensible on the merits.

The implementation is the problem, and it is leaking badly.

Tom's Hardware has direct confirmations that Opus 5.5 is now refusing reverse-engineering work unrelated to games. Its own former staffer Jarred Walton was running checks on software he is building; the work proceeded normally until 31 October, and switching to Opus 4.8 let him continue. A request to reverse engineer a product activation triggered safeguards. One Reddit user reports being suspended, then reinstated, for using Opus 5.5 with a reverse-engineering tool. Elsewhere, a studio trading as Conflict Simulations LLC says Claude Code began refusing to work on digital board-wargame modules commissioned by paying customers, and others report blocks on anti-cheat development and security reviews. Complaints are accumulating as issues on Claude Code's own GitHub repository.

Blocking anti-cheat work in the name of protecting developers is an unusually direct own goal.

It is also porous. The restriction appears confined to the combination of Claude Code and Opus 5.5, which means calling the same model through a different harness sidesteps it entirely. The first comment under Tom's Hardware's report notes simply that OpenCode doesn't have this issue. Kotaku, citing user reports, suggests the refusals now extend to IP reproduction of any kind, which Anthropic has not confirmed.

Which leaves a rule that catches legitimate work it was never aimed at and misses the behaviour it was built for, because anyone cloning a game commercially has every incentive to use a different client.

The precedent matters more than the rule. All year the question has been who governs AI in games, and the industry has offered its answers: Unity's plugin, Epic's internal mandate, Nintendo's refusal, Steam's disclosure tags. None of them is the binding constraint. A model provider changed what was buildable overnight, server-side, under a feature flag, with no opt-out and no consultation.

Seventeen days earlier, Xbox chief executive Asha Sharma had described the biggest games in the world being remixed into the source code of another as an unmistakably good outcome for players. Anthropic has now decided otherwise, for everyone using its tool, and nobody in games had a vote.