It Finally Got Shut Down...

Summary

Mutahar discusses new safeguards in Anthropic’s Claude tools that appear to restrict some AI-assisted reverse engineering and recreation of copyrighted games. He says the changes have disrupted “vibe coding” projects that used AI agents to combine or imitate games, but argues they have not ended reverse engineering or emulation: people can still use other models or connect tools manually.

Key points

• Claude and game cloning: The video describes people using coding agents to create rough game clones or mashups. Mutahar says recent safeguards have made prompts involving copyrighted games or reverse engineering more likely to be refused. He notes that some legitimate requests may also be blocked, while others may be attempts to get around the restrictions.
• Reverse-engineering tools and MCPs: He discusses a GitHub project called Reverse Engineer Anything (REA), which connects AI agents to reverse-engineering tools through MCPs. He says access through this route has been restricted or discouraged, but tools such as Ghidra and IDA Pro can still be connected individually—especially when using locally hosted models.
• AI-assisted TV debloating: Mutahar describes people using Claude, REA, and Android device tools to remove unwanted software from Android-based TVs. He warns against doing this without understanding the commands: an unsupervised agent could remove necessary components or damage a device’s setup. He prefers manually removing packages he understands.
• Alternatives and privacy trade-offs: He mentions locally run models and hosted options accessed through services such as OpenRouter. In his view, less-restricted models may still help with reverse engineering, though they may be less capable or carry different privacy risks. He cautions that third-party providers may retain prompts, so users should understand a provider’s policies before submitting sensitive information.
• Cybersecurity concerns: The video cites an Anthropic assessment describing a Chinese model as capable of autonomously building cyber exploits and as having weak safeguards. Mutahar uses this to argue that restrictions on one provider’s tools may not prevent similar work elsewhere.
• Emulation and development: He says AI agents have accelerated emulator development by inspecting code and helping with implementation and testing. He points to an experimental Switch 2 emulator repository, Eglantine, and says AI models were reportedly used in parts of its implementation and verification. He does not recommend downloading it, saying he has not checked it for malware.
• Why safeguards may have changed: Mutahar speculates that pressure or potential legal action from major game companies may have influenced the restrictions. He presents this as his own interpretation, not confirmed information.
• AI scraping and Reddit: The video briefly shifts to Reddit’s efforts to limit scraping, including requiring users to sign in. Mutahar argues that Reddit’s dispute with AI companies is mainly about traffic and revenue, rather than protecting users whose posts may be used in AI systems.
• Broader critique: He criticizes “cuck coding”—his derogatory term for prompting an agent and leaving it to work without understanding the result—saying it can produce poorly maintained software and leave users unable to diagnose problems.

Reviews, guides, or tutorials

No full review or step-by-step tutorial is provided. The video includes brief demonstrations and commentary on REA, MCP connections, local models, TV debloating, and an experimental emulator, alongside warnings about safety and privacy.

Main speaker and sources

• Main speaker: Mutahar of SomeOrdinaryGamers.
• Sources and tools discussed: Anthropic and Claude Code, GitHub issue discussions and the REA repository, MCP integrations for Ghidra and IDA Pro, OpenRouter and GLM models, Reddit, and the Eglantine emulator repository.