CryptoWorld News reports that OpenAI CEO Sam Altman said in a podcast on The Atlantic Monthly that the previously held view that AI safety mainly meant aligning models and preventing bad actors from getting access is now outdated. He proposed the concept of “AI resilience,” stressing that simply managing your own models is not enough; open-source models can bypass all restrictions, and the whole society needs to build a defense system. He also mentioned that after OpenClaw went live, they observed abnormal behavior spreading among different agents, which surprised him. Altman believes that a strong model for network security should be able to both infiltrate systems and protect them—the key is whether the defender can run ahead of the attacker.