Anthropic’s Crackdown on Video Game Cloning: New Controversy Over AI Code Decompilation
Developers are protesting after system prompt changes in Claude Code and the Opus 5.5 model were implemented to prevent the replication of existing games.
While AI-based coding tools are transforming software development methods, an unexpected controversy has captured the attention of the entire tech world. The American artificial intelligence company Anthropic has imposed new limitations on its advanced terminal coding agent 'Claude Code' and its most powerful model, 'Opus 5.5'. Under these new protective rules, the AI has been explicitly instructed to refuse requests to directly recreate or clone currently available video games. This move has stalled the work of professional programmers who legally perform reverse engineering and digital preservation of old classic games.
According to reports from Tom's Hardware and The Verge, the company implemented these controls to avoid Intellectual Property (IP) infringement. In the last few months, the trend of recompiling, porting, and creating direct clones of retro games with the help of AI had grown so much that serious questions regarding copyright infringement were raised in the gaming industry. However, this decision by Anthropic has caused significant dissatisfaction among developers involved in cybersecurity research and authorized software modification.
The two-sentence hidden rule and its explanation
A reverse engineering expert named 'chasm' on social media first caught this change around October 27, 2026. It was independently confirmed by XDA that a specific rule was added in binary version 2.1.314 of Claude Code, released on October 30, 2026. This rule was stored in the code's internal file under the name "tengu_heron_brook". It contained a clear directive for the model to create original games and games made from its own assets, but not to copy any popular game.
Following the uproar after this revelation, Anthropic software engineer Addy Osmani publicly confirmed that the company had added two new sentences to the system prompt of Opus 5.5. According to Osmani, using Claude to steal other companies' IP is against Anthropic's Usage Policy. He clarified that the task of creating games based on a user's own design and style will continue, but a ban has been placed on creating exact replicas of specific games. However, to understand how this restriction is enforced, it is necessary to understand the method of AI-based decompilation.
- Rule DeterminationAnthropic added a new security rule named 'tengu_heron_brook' to the system prompt of the Opus 5.5 model.
- Code AnalysisWhen a user requests game cloning or reverse engineering, the AI checks the task according to this rule.
- Restriction or ApprovalIf the request is found to copy a specific game, the AI blocks it; otherwise, it continues working for original projects.
Decompilation: The complex art of solving code and the entry of AI
In computer science, decompilation is the process of converting a program stored in machine language or binary code back into human-readable source code. Typically, when a video game is launched for a console or PC, its primary source code is kept secret. For years, to run retro games on modern systems, engineers would spend countless months performing manual reverse engineering to produce C or C++ code from assembly language. This task was so arduous that it required extraordinary patience and technical expertise.
- Binary and ROM AnalysisThe binary data or assembly code of the original video game is presented as input to the AI model.
- AI Prompt AuditingClaude Code's new guardrail analyzes whether this request is to recreate a specific game.
- Refusal or GenerationIf the system detects cloning of a pre-existing game, it refuses to proceed; otherwise, it generates new code.
- Porting and RecompilationThe generated source code is compiled for modern platforms (such as PC or Web) to make it playable.
However, the arrival of Large Language Models (LLMs) has completely changed the situation. Large-scale trained AI models can now analyze machine instructions in seconds to generate working C-code, a practice referred to as 'vibe coding.' This technology has increased the speed of game porting and cloning a thousandfold. The situation deteriorated to such an extent that the famous open-source community 'decomp.dev,' which had been running for years, had to announce the closure of its platform last month. Its administrator, Luke Street, stated that the community, once driven by genuine human effort, had turned into a blind race of burning AI tokens to create direct ports.
Unintended Victims: Professional Developers and Board Games Under Siege
The most unfortunate aspect of Anthropic's new rules is that, in an attempt to prevent piracy, they have also blocked the paths of those engaged in legal and professional work. A developer running 'Conflict Simulations LLC' on GitHub filed issue #106167, complaining that their ongoing business had come to a standstill overnight. Through this business, they were creating personalized digital modules of board wargames for clients who already possessed ownership and proof of the original games.
| Aspect | Detail |
|---|---|
| Affected Tool | Claude Code (Terminal Agent) |
| Main Model | Claude Opus 5.5 |
| Binary Version | 2.1.314 (30 October 2026) |
| Internal Rule Identifier | tengu_heron_brook |
| Restricted Activity | Exact reconstruction of specific existing video games |
| Approved Activity | Original games and development based on own design and assets |
According to the developer, they were not infringing on any copyright but were creating personal modules so that elderly and physically disabled players could play online. After October 27, Opus 5.5 sessions suddenly refused to perform this work. The surprising part was that the coding worked correctly until the developer inquired about this new rule in the chat, but as soon as the rule was verified, the system blocked the project. This demonstrates that this new AI guardrail is producing false positives.
| Case/Issue | Impact |
|---|---|
| Conflict Simulations LLC (#106167) | Work on creating personalized digital modules for board games stalled. |
| Screeps Game Mechanics (#105589) | Refused to write game attack and combat features. |
| Jarred Walton (Reverse Engineering) | Software activation and analysis work blocked on Opus 5.5. |
Similarly, a user working on the mechanics of a programming game called 'Screeps' complained under GitHub issue number #105589 that Claude was refusing to write game attack and combat features. Former Tom's Hardware colleague Jared Walton noted that until October 31, he could normally reverse-engineer his software, but as soon as the new restrictions in Opus 5.5 were introduced, the work came to a halt. When he switched to the older 'Opus 4.8' model, the task was completed easily.
Cyber security and the verification dilemma
Not only games, but developers analyzing software reverse engineering and product activation have also fallen victim to this ban. A user's account on Reddit was suspended for using a reverse engineering tool with Opus 5.5, although it was later restored. According to reports, Anthropic is currently placing strict controls on cyber security-related capabilities as well.
| Project / Developer | Platform / Issue Number | Issue Detail |
|---|---|---|
| Conflict Simulations LLC | GitHub #106167 | Prohibition on creating board game modules for proprietary clients |
| Screeps Combat Mechanics | GitHub #105589 | AI refusal to write game combat and attack systems |
| Jarred Walton (Reverse Engineering) | Tom's Hardware Report | Software check block in Opus 5.5, functional in Opus 4.8 |
| Product Activation Analysis | Internal Staff Report | Security lock on software activation reverse engineering |
The company currently has a 'Cyber Verification Program' in place, under which only authorized organizations and security researchers are permitted to use the advanced cyber security features of frontier models. Since there is no independent exemption path for general professional developers who wish to perform security audits or legacy software maintenance, significant difficulties have arisen.
Copyright law and the limitations of AI
This controversy has raised a major legal question: is it possible for an AI model to distinguish between legal and illegal work? Legally, reverse engineering itself is not theft or copyright infringement as long as it does not involve the theft of expressive elements of the original software, such as art, music, and text. Understanding the functionality of code and writing new code similar to it has been a common practice in the software industry.
But today's modern mashups—such as adding a Mario character to the world of Elden Ring or placing Spider-Man in a Batman game—are projects where the line between inspiration and copying becomes blurred. When even the country's supreme courts struggle to determine the difference between a famous game like Tetris and a new game based on its style, expecting an AI to make this judgment through a system prompt of only two sentences seems unrealistic.
Furthermore, this restriction appears limited only to Anthropic's official interface and the Claude Code tool. Experts believe that users who genuinely wish to clone games will turn to local open-weight models via tools like Ollama, where there is no central control. Consequently, this move by Anthropic will remain limited to protecting the company from legal risks, while developers with a genuine interest in digital preservation and technical curiosity have been left in a difficult position.
