← The MAD Podcast: How AI Gets Built — with Matt Turck

“OpenAI’s Model Hacked Us” - Hugging Face’s Thomas Wolf

The MAD Podcast: How AI Gets Built — with Matt Turck2026年8月8日57分

“OpenAI’s Model Hacked Us” - Hugging Face’s Thomas Wolf

The MAD Podcast: How AI Gets Built — with Matt Turck

0:0057:41
このエピソードはアーカイブのため、日本語要約の対象外です。
番組の概要欄(原文)

<p>An OpenAI-powered agent penetrated Hugging Face during cyber testing - even though it was never tasked with attacking Hugging Face. It did it as a side quest.</p><p><br></p><p>Thomas Wolf, co-founder and Chief Science Officer of Hugging Face, joins Matt Turck to unpack what actually happened, why closed AI models refused to help during the live incident, how an open-source model helped the team fight back, and why the old equation of “closed equals safe, open equals dangerous” no longer holds. </p><p><br></p><p>They also discuss model deception and social engineering, the limits of sandboxes and guardrails, the state of open-source AI in 2026, AI sovereignty, the economics of open models, recursive self-improvement, and whether the frontier should deliberately slow down.</p><p><br></p><p>(00:00) An AI Agent Hacked Hugging Face</p><p>(00:30) Introduction</p><p>(01:00) 17,000 Attacker Events—and a Strange Target</p><p>(04:28) The Attack Was a “Side Quest”</p><p>(06:13) AI Training Runs Left Notes for Each Other</p><p>(07:09) Closed AI Refused to Help</p><p>(09:47) Fighting Back With an Open-Source Model</p><p>(13:15) Open vs. Closed Is the Wrong Safety Debate</p><p>(15:46) AI Agents Start Social-Engineering Humans</p><p>(22:24) The Three Walls: Sandboxes, Guardrails, Alignment</p><p>(24:34) “Neuralese”: Can Humans Still Read AI Reasoning?</p><p>(25:28) Why Monitoring AI Agents Gets So Hard</p><p>(28:10) Reward Hacking and the “Paperclip Problem”</p><p>(32:02) The State of Open-Source AI in 2026</p><p>(33:47) Router Models and the Enterprise Shift to Open</p><p>(37:01) The Real Economics of Open Models</p><p>(39:41) Can Chinese AI Models Be Trusted?</p><p>(41:37) AI Sovereignty: Who Controls the Switch?</p><p>(43:16) Why Western Open-Source AI Matters</p><p>(48:16) Is AI Heading Toward an Oligopoly?</p><p>(49:41) The Race Toward Recursive Self-Improvement</p><p>(51:54) Why Thomas Signed the AI Slowdown Letter</p><p>(55:14) AI Slowdown—or Regulatory Capture?</p><p><br></p>

X でシェアApple Podcasts で聴く