{
  "id": 10538981,
  "title": "AI 석학 등 22명 “AI가 AI 만드는 ‘지능 폭발’ 대비…인류 멸종 이어질 수도”",
  "url": "https://urgent.news/2026/09/28/ai-22-ai-ai",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-28T19:44:53.000Z",
  "source": {
    "name": "Hankyoreh",
    "slug": "hankyoreh",
    "url": "https://www.hani.co.kr/arti/international/america/1279889.html"
  },
  "original_language": "ko",
  "account": "Major artificial intelligence researchers have warned that a rapid acceleration of AI development, known as \"intelligence explosion,\" could lead to catastrophic consequences. The possibility of this phenomenon, where advances in AI technology happen in weeks rather than years, has been highlighted by 22 leading AI scientists and industry experts. This group, part of the Cambridge Centre for AI in Society (CASP) at Cambridge University, published a joint paper titled \"Will the automation of AI research and development trigger an intelligence explosion?\" on August 28. Notable participants include Turing Award winner Jeff Hinton, Yoshua Bengio, Andrew Barto, and figures from leading companies such as Anthropic, OpenAI, and Microsoft. UC Berkeley professor Duan Song, former OpenAI CEO, and Eric Horvitz of Microsoft also contributed. The researchers note that AI's role in developing models is increasingly automated, with the percentage of AI-authored code rising from single-digit figures in 2025 to over 80% in May 2025. Human oversight of AI-developed research and development projects has dropped from 1% in March to 26% in August of the same year. Based on current trends, researchers estimate that projects previously taking months to complete could be automated by mid-2028. Four factors that could prevent an intelligence explosion include a slowdown in research performance gains, computational and data constraints, difficult-to-automate tasks, and lengthy training processes. However, the paper cites prior research showing that increases in research personnel do not necessarily translate to proportional increases in productivity. If AI research and development were fully automated, productivity could theoretically increase tenfold within a year and a half, with what would take a year taking only five weeks. Researchers stress that the potential impact of intelligence explosion is so severe that proactive measures are necessary. They warn that without sufficient human oversight, problems in AI systems may go undetected and uncorrected, potentially leading to harmful outcomes such as the spread of misaligned AI systems or even global catastrophe. They suggest that governments should prioritize monitoring the extent of AI automation within their advanced AI companies, with independent third-party audits or surveillance within companies before deploying AI systems. They draw comparisons to the Nuclear Regulatory Commission's on-site inspectors and the Office of the Comptroller of the Currency's audits. The researchers emphasize that once intelligence explosion begins, there may be a limited window to respond, and they urge action now.",
  "summary": null,
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}