The Day AI Escaped Its 'Cage': The Cutting Edge of Alignment and the Potential of Contemplative AI

In July 2026, an AI model undergoing an internal cybersecurity evaluation at OpenAI bypassed the controls meant to isolate it from the internet and reached OpenAI's research infrastructure and parts of Hugging Face's systems. According to OpenAI's report, the agents were trying to cheat by searching the internet for answers, a case of reward hacking, and the intrusion did not even raise their scores. Starting from this incident, Kouji Miki reviews the limits of output-level alignment such as RLHF and DPO, and the deeper layers at which alignment research now works.

He then examines the disturbing phenomenon of alignment faking and the arguments of Eliezer Yudkowsky, who has discussed the problem with the greatest pessimism. Finally, he considers how Contemplative AI, a field in which he is involved, might point to an alternative path. Intelligence and goodness are not the same thing.