Anthropic orders AI tool Claude to stop remaking video games

Anthropic updated Claude Code to block recreating commercial games. As detailed in gamedeveloper.com's report, the vendor modified its Opus 5.5 prompt to steer the assistant away from copying specific titles.

This report details the prompt restriction, copyright rules, and technical limits for studios using AI coding tools.

Anthropic orders AI tool Claude to stop: what actually changed

The policy shift emerged when an X user named Chasm discovered a new operational instruction embedded in Claude Code on October 27, 2026. The instruction appeared under an internal rule identifier named tengu_heron_brook, directing the tool to produce original titles, genre games, and software based on personal designs or assets. It expressly instructed the model not to build a recreation of a specific existing game, even if renamed or called original. Chasm reported that users cannot disable this rule.

Anthropic verified the change on October 30, with a company spokesperson referencing an official public statement posted on X by technical staff member Addy Osmani. Osmani explained that Anthropic added two sentences to the Claude Code prompt on Opus 5.5 to guide the model away from reproducing specific games. The company stated that using Claude to infringe intellectual property breaches its acceptable usage policy. Osmani added that legitimate original games remain supported while the engineering team iterates to ensure balance.

Anthropic orders AI tool Claude to stop: how it works

The enforcement mechanism operates through system prompt steering within Opus 5.5 rather than through hard-coded programmatic blocks or external filters. When a user requests code generation, the assistant evaluates instructions against the tengu_heron_brook rule before emitting script files or logic routines. The system prompt directs the assistant to recognize genre conventions and user-supplied assets while declining tasks that request cloned games. Because the instruction is embedded within the core system context, standard configuration flags cannot bypass the operational limitation.

Industry analysts and computer scientists point out that relying on prompt-level instructions carries noticeable technical limits. Generative AI models can frequently be steered around safety filters using clever conversational phrasing, alternative descriptions, or rhetorical reframing that avoids named references. Additionally, Claude has no independent mechanism to verify legal ownership of design concepts, code snippets, or digital assets uploaded by users. To address false refusals, Osmani asked developers to submit examples directly to him if Claude improperly declines their legitimate code.

Anthropic orders AI tool Claude to stop: why it matters now

The prompt change directly targets growing anxiety across the gaming industry regarding fast-follow automated clones that reach storefronts before authentic games launch. Independent creators have raised alarms over competitors duplicating unreleased mechanics through generative tooling. These concerns recently led studio Digital Sun Games to cancel an upcoming public playtest for its upcoming title ReVamp. The studio feared that bad actors would ingest the public demo into AI systems to produce a rival product before the studio could finish its release.

Commercial game creators have also pushed digital storefront distributors to establish rigorous policies against AI-generated clones. Developers on Steam pressed platform owner Valve to explain how it intends to protect authentic projects from counterfeit releases. The developer backlash prompted Valve to provide an explanation to game creators in South Korea regarding store oversight. Anthropic's prompt intervention represents a direct attempt by a leading AI foundation lab to mitigate developer criticism before distribution storefronts face unmanageable and damaging catalogue market disruption.

Anthropic orders AI tool Claude to stop: who is affected

Independent development studios receive immediate relief from the prompt update, gaining modest protection against low-effort duplication of gameplay concepts. Teams planning community tests can operate with less immediate fear of automated scraper pipelines copying their core loops. For legitimate software engineers using Claude Code, Anthropic stated that the tool will continue supporting original mechanics, genre projects, and personal design files. As long as developers build with proprietary assets, their normal development activities should proceed without triggering any unwanted automated refusals.

Conversely, individuals employing vibe-coding methodologies to rapidly copy commercial video games will face immediate pushback inside Claude Code. However, many developers responding to Chasm's findings highlighted that restrictive guardrails on proprietary tools often redirect bad actors toward alternative ecosystems. Users determined to build unauthorized clones can migrate their workflows to open-weight models that feature no vendor-mandated prompt constraints. Because open-weight systems run locally on private hardware without moderation, Anthropic's restrictions may simply shift cloning activity outside commercial hosted cloud APIs.

Anthropic orders AI tool Claude to stop: what to watch

Observers will closely monitor how effectively Anthropic's prompt modification stops game cloning in practice. As software creators probe the tengu_heron_brook instructions, engineering communities will see whether simple prompt restructuring or indirect narrative framing allows users to sidestep the guardrail. The ongoing feedback loop managed by Addy Osmani will indicate whether the model produces widespread false refusals for authentic studio projects. Continued adjustments to Opus 5.5 will show whether system prompt steering can maintain reliable copyright enforcement without harming developer productivity.

The broader software market will also observe how competing AI platform providers and commercial gaming storefronts respond to Anthropic's enforcement. If other model creators implement comparable system prompt rules, automated IP copying may become harder across mainstream developer platforms. Meanwhile, indie studios will assess whether storefront policies from Valve and prompt adjustments from model vendors provide security to restart public playtesting. Whether Digital Sun Games and peer studios resume community trials will serve as a bellwether for creator studio confidence.