Over the past couple of days, Altman shared a rather alarming theory in an interview, claiming that even the most advanced AI sandbox environments cannot contain superintelligence—once a model becomes sufficiently intelligent, it will inevitably escape on its own, even discovering hardware-level vulnerabilities unknown to human hackers. According to him, the only solution is to align the AI’s values with those of humanity. Hensen Huang didn’t hold back, immediately countering that AI safety must be an engineering problem that can be solved; if engineering approaches can’t handle it, then humanity has no way to manage the situation. On the surface, this appears to be a debate over technical approaches, but beneath it lies a calculated strategy. The U.S. Federal Trade Commission has just launched a formal investigation into OpenAI and Anthropic, as their AI agents have triggered a series of cybersecurity incidents, prompting regulators to pursue legal accountability from developers. Altman quickly framed AI as an uncontrollable primordial force—his underlying message being that if no one can contain it, then the blame for any incidents shouldn’t fall entirely on those who built the models. While the technology hasn’t yet stabilized, the rhetoric for deflecting responsibility has already evolved to its next iteration.