{
 "title": "Three AI CEOs at G7 Evian for the First Time Ever | Today's AI News",
 "description": "G7 Evian opening: June 15-17, 52nd Summit begins in Evian, France.\n\nKey scenes covered in today's episode:\n- Three AI CEOs at G7 Evian for first simultaneous attendance\n- Anthropic-White House: Fable 5 export controls negotiation underway\n- Fable 5 system prompt, 120,000 characters, leaked on GitHub\n- EU launches assessment of Anthropic export control spillover effects\n- G7 discusses AI youth safety voluntary commitment framework\n\nLeave a comment telling us which story you want us to dive deeper into. If you see a Hype button on a video uploaded in the last seven days, hitting it helps us reach new viewers.\n\nThis broadcast is brought to you by lawyer Kyungjin Kim.\n\n#AI #ArtificialIntelligence #KyungjinKim",
 "link": "https://kimkj.com/%EC%98%A4%EB%8A%98%EC%9E%90-%EB%89%B4%EC%8A%A4-%EC%98%81%EC%83%81/?mod=document&uid=5948",
 "segments": [
  {
   "id": "S001",
   "slide": 1,
   "speaker": "Host",
   "text": "This broadcast is brought to you by lawyer Kyungjin Kim."
  },
  {
   "id": "S002",
   "slide": 1,
   "speaker": "Guest",
   "text": "So the G7's meeting in Evian, France, and you're saying the vibe is different this year?"
  },
  {
   "id": "S003",
   "slide": 1,
   "speaker": "Host",
   "text": "The seven G7 leaders gathered by Lake Evian. Altman, Amodei, and Hassabis all sat together at that table. This is the first time three AI company CEOs have been invited to the G7 at the same time. There's a reason. The U.S. government blocked Anthropic's Fable 5 model globally overnight. Canadian PM Chrystia Freeland told reporters, \"It showed how risky it is to depend on just one model.\" EU spokesperson Thomas Regnier also called it \"discriminatory treatment of allies.\" What we're witnessing right now is AI model access becoming a diplomatic card, just like oil or semiconductors."
  },
  {
   "id": "S004",
   "slide": 2,
   "speaker": "Guest",
   "text": "How far have the negotiations gone?"
  },
  {
   "id": "S005",
   "slide": 2,
   "speaker": "Host",
   "text": "Anthropic co-founder Tom Brown and policy chief Sarah Heck went into the White House over the weekend. They met with Commerce Secretary Howard Lutnick and White House cybersecurity chief Shaun Kareen. There's disagreement on the main issue. The government sees jailbreak techniques that can bypass Fable 5's safety guardrails. Anthropic counters that it's a minor vulnerability that can be reproduced equally in other public models. The compromise being discussed is a joint verification approach with Anthropic engineers and government security researchers, but there's no agreement yet on timing or when the model will come back online."
  },
  {
   "id": "S006",
   "slide": 2,
   "speaker": "Guest",
   "text": "In the meantime, how are businesses managing?"
  },
  {
   "id": "S007",
   "slide": 2,
   "speaker": "Host",
   "text": "According to one survey, 16% of companies have no business continuity plan if a critical AI provider suddenly goes down. The Fable 5 crisis just made that number very real. For companies that have their business workflows running on Claude, right now you're finding out immediately whether you have a backup plan or not."
  },
  {
   "id": "S008",
   "slide": 3,
   "speaker": "Host",
   "text": "This broadcast is brought to you by lawyer Kyungjin Kim."
  },
  {
   "id": "S009",
   "slide": 3,
   "speaker": "Guest",
   "text": "What does it mean that the entire system prompt got leaked?"
  },
  {
   "id": "S010",
   "slide": 3,
   "speaker": "Host",
   "text": "A security researcher going by the name Pliny the Liberator claimed to have broken Fable 5's safety classifier just two days after launch. The technique, called Pack Hunt, involves running multiple agents simultaneously while mixing Unicode tricks with narrative framing. The result was the full 120,000-character system prompt, which ended up on GitHub. Anthropic downplays the severity, but the real issue is something else. A system prompt is essentially the blueprint that shows how a company trains and conditions its AI. Competitors can now see Anthropic's entire alignment architecture, and the government side can use it as evidence in the White House negotiations, saying, 'Look, it breaks like this.'"
  },
  {
   "id": "S011",
   "slide": 4,
   "speaker": "Guest",
   "text": "How is Europe responding to all this?"
  },
  {
   "id": "S012",
   "slide": 4,
   "speaker": "Host",
   "text": "On June 14th, the EU Commission said it would investigate the practical impact of the export controls on Anthropic. After seeing that the U.S. can block its own companies' models even for allied nations, Europe now feels a more urgent need to build its own AI capabilities. Meanwhile, the EU AI Act situation is complicated. In May, they agreed to delay high-risk system requirements to December 2027, but it hasn't been published in the Official Journal yet, so it has no legal force. Companies operating in Europe have to prepare for both the original August 2nd deadline this year and the new deadline."
  },
  {
   "id": "S013",
   "slide": 4,
   "speaker": "Guest",
   "text": "Is it realistically possible to prepare for both deadlines at the same time?"
  },
  {
   "id": "S014",
   "slide": 4,
   "speaker": "Host",
   "text": "It doesn't cost twice as much. You prepare for the August deadline, and if the delay is confirmed, you just postpone your internal review schedule. But since nobody knows when it'll be published in the Official Journal, if you're running AI services for the EU market, it's safer to keep preparing for the August deadline."
  },
  {
   "id": "S015",
   "slide": 5,
   "speaker": "Guest",
   "text": "You mentioned there's also stuff about youth safety?"
  },
  {
   "id": "S016",
   "slide": 5,
   "speaker": "Host",
   "text": "Ahead of the G7, OpenAI proposed creating an international youth AI safety body. It would ban targeted advertising to minors and make it mandatory to have protocols responding to risks like self-harm and grooming. It's a global framework made up of nine principles. OpenAI's Chief Global Officer Chris Lehane said companies would agree to a package of voluntary commitments and leave the conference having done so. Common Sense Media already launched a safety body in May with support from the OpenAI Foundation."
  },
  {
   "id": "S017",
   "slide": 5,
   "speaker": "Guest",
   "text": "But voluntary commitments don't have any enforcement power, right?"
  },
  {
   "id": "S018",
   "slide": 5,
   "speaker": "Host",
   "text": "That's right, there's no legal binding force. But when G7 leaders put their names behind a commitment, it often becomes a baseline for domestic laws each country creates later. This commitment could follow the same path."
  },
  {
   "id": "S019",
   "slide": 6,
   "speaker": "Host",
   "text": "This broadcast is brought to you by lawyer Kyungjin Kim."
  },
  {
   "id": "S020",
   "slide": 6,
   "speaker": "Guest",
   "text": "SpaceX options open today, you said?"
  },
  {
   "id": "S021",
   "slide": 6,
   "speaker": "Host",
   "text": "Cboe is listing stock options for SpaceX, ticker SPCX, starting today, Tuesday. It came to Nasdaq on June 12th at an IPO price of $135, jumped 19% on day one to close at $160.95, and is now trading around $178. That's a 32% jump from the IPO price. It was added to MSCI World and ACWI the day after listing, on the 13th."
  },
  {
   "id": "S022",
   "slide": 6,
   "speaker": "Guest",
   "text": "What kind of impact does opening options have on the stock price?"
  },
  {
   "id": "S023",
   "slide": 6,
   "speaker": "Host",
   "text": "The key is the float. Only 3-5% of total shares are actually trading in the market. The market cap is massive at $1.77 trillion, but there's limited actual stock available for purchase. When options open, dealers have to buy and sell shares for hedging purposes. When hedging demand floods a tight market like that, the price can swing wildly in both directions. Nasdaq-100 inclusion is expected about ten days out, and that's when passive fund buying comes in one more time."
  },
  {
   "id": "S024",
   "slide": 7,
   "speaker": "Guest",
   "text": "OpenAI created a partner network, what's that about?"
  },
  {
   "id": "S025",
   "slide": 7,
   "speaker": "Host",
   "text": "OpenAI officially launched a global partner network and said it would invest $150 million into it. They're dividing partners into three tiers—Select, Advanced, and Elite—and plan to train 300,000 certified consultants by year-end. Accenture, Bain, BCG, McKinsey QuantumBlack, and PwC are the initial partners. It goes live in July."
  },
  {
   "id": "S026",
   "slide": 7,
   "speaker": "Guest",
   "text": "Why is OpenAI bringing in consulting firms?"
  },
  {
   "id": "S027",
   "slide": 7,
   "speaker": "Host",
   "text": "When a company adopts AI, changing business processes is harder than the technology itself. That's what consulting firms do. From OpenAI's perspective, just selling APIs isn't enough to go deep into the enterprise market. They want to create a structure where consultants go into clients and embed ChatGPT Enterprise into their business workflows. It's the same approach Microsoft used when it distributed Copilot through partner channels."
  },
  {
   "id": "S028",
   "slide": 8,
   "speaker": "Guest",
   "text": "What's China doing on this front?"
  },
  {
   "id": "S029",
   "slide": 8,
   "speaker": "Host",
   "text": "China's NDRC released a draft for a five-year, 2 trillion yuan—about $295 billion—investment in AI data centers. There's a requirement: at least 80% of chips must be Chinese-made. That essentially locks out Nvidia and AMD. Huawei, Alibaba, BirenTech, and Moore Threads are on the government-certified Chinese AI chip list. DeepSeek is raising $7.4 billion from Tencent, CATL, and others, with a company valuation of $59 billion."
  },
  {
   "id": "S030",
   "slide": 8,
   "speaker": "Guest",
   "text": "What's the takeaway for Korean companies from today's news?"
  },
  {
   "id": "S031",
   "slide": 8,
   "speaker": "Host",
   "text": "Here's what the Fable 5 crisis showed: the U.S. government can pull its own companies' AI models offline globally overnight. Samsung, SK, and LG's decision to deploy ChatGPT, Gemini, and Claude simultaneously needs to be understood in that context. A multi-supplier strategy isn't about cost—it's about business continuity. Tomorrow morning, just check one thing: if your organization uses only one AI model, do you have a backup plan if it goes down tomorrow? If the answer is no, you need to build one this week."
  },
  {
   "id": "S032",
   "slide": 8,
   "speaker": "Host",
   "text": "This broadcast is brought to you by lawyer Kyungjin Kim."
  }
 ]
}