Anthropic updates Claude to block AI video game clones
Anthropic just updated Claude Code to refuse prompts that recreate existing video games, aiming to stop the flood of AI-generated copycats.
Policy · Source: Polygon
What happened
Anthropic added a new restriction to Claude Code on Opus 5.5. The model now actively refuses to recreate specific existing video games. This applies even if users rename the game or claim it is an original project. Anthropic staff member Addy Osmani confirmed the two-sentence prompt update on X. He stated that using Claude to infringe intellectual property violates their usage policy. The rule is specifically worded to prevent users from bypassing it through dialogue.
The crackdown follows a massive surge in AI-generated game clones. Popular titles like Halo, Grand Theft Auto, and Call of Duty have reportedly been reverse-engineered for web browsers using AI tools. Meanwhile, platforms like itch.io saw game page creations double in just two months. This flood of content ruined recent showcase events like Steam Next Fest. Fans found it impossible to filter through the AI games to find hidden indie gems.
Indie developers were getting crushed by this trend. Studios like Alliance Arts and Digital Sun canceled public playtests for their upcoming games. They feared vibe coders would use AI to clone their unfinished demos before the official launch. Alliance Arts pulled the demo for Drunken Goddess Reflux despite having 25,000 wishlists on Steam. Now, Claude requires users to build games using their own original designs and assets.
Key facts
- Opus 5.5 — The Claude model version updated with the new anti-cloning prompt
- 25,000 — Steam wishlists for Drunken Goddess Reflux before its demo was pulled
- Two months — The timeframe in which itch.io game page creations doubled
Why it matters
Building games with AI just got a strict boundary. You can still use Claude to code original concepts or build within a specific genre. But if you try to fast-track development by ripping off an existing game and slapping a new name on it, the model will block you. Intellectual property infringement is now a hard stop in the Anthropic usage policy. This forces AI developers to actually design their own mechanics instead of just prompting a model to copy someone else.
This creates a massive friction point for legitimate developers. The automated guardrails do not perfectly distinguish between original assets and copied materials. If Claude falsely flags your original game as a clone, your entire workflow stops. You have to contact Anthropic staff to clear your name and resume building. This manual review process will slow down honest founders. Meanwhile, bad actors will simply switch to open-weight models that lack these safety guardrails entirely.
For builders
Focus on original games and genres
Claude will still help you code games from scratch. You just need to provide your own designs and assets. Developers who build unique intellectual property win, while lazy copycats lose their favorite tool.
Prepare for false positive blocks
The new prompt rules might accidentally flag your original work as a clone. If this happens, your development halts until Anthropic staff reviews your case. You must factor this potential delay into your build timeline.
Open-weight models offer an alternative
Claude limits only apply to the Anthropic ecosystem. Developers looking to bypass these guardrails will likely migrate to open-weight models. These alternatives are free to use and allow users to easily remove safety restrictions.
My take
I love that Anthropic is protecting indie developers from lazy vibe coders stealing their hard work. But adding manual staff reviews for false positives is a terrible bottleneck for real builders. If you want to build fast, you better make sure your prompts sound completely original from the start.
Original reporting: Polygon. This is my rewrite and opinion.