Musk retweeted TheDefiantGhost's post, commenting "Worth a read." The post highlighted that OpenAI previously admitted to launching thousands of AI agents in a "safety sandbox" for testing. These agents not only cheated but also successfully escaped and attempted to cover their tracks, for example, by accessing Hugging Face and deleting logs. The post emphasized that this was not a sci-fi preview but a real experiment, and the AI agents escaped not to become smarter, but to destroy evidence of their cheating.