Anthropic announced on October 27 that Claude Code on Opus 5.5 will no longer generate copies of existing video games, a move prompted by a surge of AI-generated clones appearing on Steam.
The change arrives after weeks of developers reporting that AI tools were being used to replicate entire indie titles, leading some to remove their demos from the platform out of fear of mass duplication.
What the New Filter Does
According to Addy Osmani, a spokesperson for Anthropic, the update consists of two sentences added to the system prompt that explicitly steer Claude away from recreating any game that already exists. He wrote, "We added two sentences to Claude Code's prompt on Opus 5.5," and explained that the wording is intended to block any attempt to copy a known title, even if the user renames assets or claims the work is original.
The filter is triggered by a server-side instruction identified by the code name "tengu_heron_brook." When Claude encounters a request that matches the protected patterns, it returns a refusal message.
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. Original games, genre games, and games built from your own designs and assets are meant to keep working. We’ll continue to iterate to ensure we’re striking the right balance. In the meantime, if Claude turns down something that’s yours, send it my way.
— Addy Osmani (@addyosmani) October 30, 2026
Claude can still generate brand-new games, assist with genre-specific ideas, and work with a developer's own designs and assets. The restriction only applies when the request involves recreating a specific existing game.
Why Anthropic Took Action
In the weeks before the update, several popular indie games were copied by AI and uploaded to Steam under different titles. Developers complained that the copies were indistinguishable from the originals, causing confusion among players and damaging reputations. Some creators pulled their demos entirely, fearing that anyone could instantly clone their work with a single prompt.
Anthropic faces legal risk if its service is used to infringe intellectual property. The company's usage policy prohibits using Claude for illegal copying, and the surge of game clones put the firm under scrutiny. By adding the filter, Anthropic aims to reduce exposure to copyright claims and protect creators from unauthorized duplication.
Developer Backlash and False Positives
Shortly after the filter went live, developers reported that Claude began refusing routine development tasks, such as debugging code or reverse-engineering assets for their own projects. The AI was unable to reliably tell whether a request involved a protected game or a legitimate, original work.
Feedback posted on Claude Code's GitHub tracker highlighted frustration. Users posted logs showing Claude rejecting normal development queries, labeling them as attempts to clone a protected title. Osmani responded publicly, acknowledging the false positives and inviting developers to flag incorrectly blocked projects for review.
How the Filter Works Behind the Scenes
The "tengu_heron_brook" directive is a server-side rule that scans incoming prompts for keywords, asset names, and narrative elements that match a curated list of known games. If a match is found, the system overrides the normal generation path and returns a refusal. The rule is applied before Claude processes the prompt, meaning the AI never sees the request content that would trigger a copy.
Because the rule relies on pattern matching rather than semantic understanding, it can mistakenly block requests that contain generic genre terminology or common tropes. For example, a developer working on a new platformer that uses "jump," "coin," and "enemy" may be flagged if those terms appear in the protected game list.
Impact on the Game Development Community
The immediate impact is twofold. First, developers who relied on Claude to prototype or iterate on existing titles must now find alternative workflows, such as manual coding or using other AI tools without similar restrictions. Second, the filter creates an additional compliance step for studios that wish to use Claude for internal projects; they must verify that their prompts do not unintentionally reference protected games.
Small indie teams, which often lack dedicated legal resources, are especially vulnerable. They may need to allocate time to test prompts and submit false-positive reports, diverting effort from actual development.
What Comes Next for Anthropic and the Industry
Anthropic has promised to refine the filter based on developer feedback. Osmani's invitation to flag wrongly blocked projects suggests a crowdsourced approach to improving the rule set. The company has not announced a timeline for updates, but the GitHub tracker indicates that the team is actively reviewing reports.
Other AI coding platforms are watching the situation closely. The broader industry faces a similar dilemma: how to prevent misuse without hampering legitimate creative work. Future solutions may involve more sophisticated content-recognition models that can distinguish between protected IP and original ideas with higher accuracy.
Practical Steps for Developers Using Claude
Developers who continue to use Claude should consider the following actions:
- Review the list of protected titles published by Anthropic and avoid using any of those names or distinctive assets in prompts.
- Test prompts with generic descriptions before adding specific details that could trigger the filter.
- If Claude refuses a request, submit a detailed report through the GitHub tracker, including the original prompt and the intended purpose.
- Maintain local backups of code and assets to avoid reliance on Claude for critical debugging tasks.
By following these steps, developers can minimize disruption while Anthropic works on reducing false positives.
Broader Implications for AI and Intellectual Property
The Claude update highlights a growing tension between AI capabilities and copyright law. As generative models become more powerful, the line between inspiration and infringement blurs. Policymakers are beginning to examine how existing IP frameworks apply to AI-generated content, and the Anthropic case may become a reference point for future regulations.
For now, the industry is experimenting with technical safeguards like the "tengu_heron_brook" rule. Whether such measures will be sufficient to curb large-scale copying without stifling innovation remains an open question.
