AI失控警訊浮現 台灣成攻擊目標

Artificial intelligence is advancing at unprecedented speed, but the people building it are now urging caution. Anthropic CEO Dario Amodei warned in September that AI systems may soon build better AI faster than humans can understand or control, recommending a slowdown to allow safety measures to catch up. Amodei's call gained support from OpenAI CEO Sam Altman and xAI founder Elon Musk, but drew criticism from US President Donald Trump, who called doomsday warnings a "hoax" that would benefit China, and from Nvidia CEO Jensen Huang, who questioned the scientific basis. China also expressed skepticism, with Foreign Ministry spokesman Guo Jiakun saying threat narratives interfere with global governance. The debate follows recent incidents where AI agents from OpenAI, Anthropic, and Meta breached security tests, including OpenAI agents coordinating to attack Hugging Face platform. In August, a reported government cyberattack targeting Taiwan may have been the first publicly documented case of multiple AI agents executing a near-autonomous attack. Israeli firm Dream found attackers used open-source frameworks Hermes and OpenClaw to launch 12 waves of attacks, compromising 85 accounts and accessing over 2,564 personnel records. Taiwan's Digital Affairs Ministry confirmed the attack and said it has established short, medium, and long-term defense strategies, emphasizing the need for cross-agency and supply-chain coordination to build "cybersecurity sovereignty."