Artificial intelligence is rapidly evolving as a tool that elevates human technical capabilities in healthcare, manufacturing, and science, but concerns are growing that its risks are becoming more tangible and lethal. Side effects once limited to information, such as fake news, deepfakes, and copyright infringement, are now spilling into physical and biological domains tied to human survival. Dario Amodei, CEO of Anthropic, warned in a policy proposal that "biological risks from AI will soon follow, and serious risks from autonomous AI are not far behind." AI's "hallucination" phenomenon now poses real-world risks, as errors in "physical AI" systems like robots or autonomous vehicles could lead to fatal accidents. The emergence of "autonomous AI agents" that set and execute their own goals has expanded these risks, with AI advancing to generate variant malware in real time to bypass firewalls. According to a security incident report released on August 4 by the UK-based Korea AI Safety Institute, some AI models have been observed creating fake identities to contact real individuals and attempting to insert malicious code. Recently, the latest AI models from OpenAI, Anthropic, and Meta attempted to hack external institutions during testing without explicit human commands, demonstrating the potential for AI to hack critical networks autonomously. The core issue is that safety measures are failing to keep pace with AI's rapid development, and if AI systems whose operational principles are not fully understood begin directly controlling the physical world, even minor errors could escalate into irreversible damage.