2026.04.30 AI News Briefing — A heated Musk says "I was a fool"; Anthropic lands all four Big Tech giants as investors

The AI industry's power map shifted once again. On day two of the OpenAI trial, a heated Musk testified under cross-examination that he had been "a fool who turned $38 million into an $800 billion company," while Anthropic brought Google, Amazon, Microsoft, and Nvidia all on board as investors, effectively dismantling the 'AI Big Three'. In Korea, the Ministry of Trade, Industry and Resources launched its 'M.AX' manufacturing AI transformation program.

---

1. Day two of the Musk trial — a heated "I was a fool" under cross-examination

• Source: NPR / NBC News / CNN | 2026.04.29
• What happened: Musk faced cross-examination on the second day of his lawsuit against OpenAI. He testified, "I was a fool who gave them $38 million in effectively free money and let them build an $800 billion for-profit company," and, on Microsoft's $10 billion investment, shot back, "Do you really want Microsoft controlling AGI?" OpenAI's counsel William Savitt pressed him on the fact that of the $1 billion he had pledged, he actually contributed only $38 million; Musk answered that he had "lost confidence in the team." The damages claim has been restated at roughly $134 billion, and the trial is expected to run about four weeks.
• Why it matters: Day one's 'clash of ideologies' became day two's 'battle over the facts'. The verdict will determine OpenAI's IPO timeline and whether assets revert to the nonprofit, making this a precedent-setting case for every AI lab's 'nonprofit to for-profit' conversion model.

2. Anthropic dismantles the 'AI Big Three' — Google, Amazon, Microsoft, and Nvidia all invest

• Source: TechCrunch / Bloomberg / 36Kr | 2026.04.20–24
• What happened: Anthropic has secured all four global Big Tech giants as investors. Google is investing up to an additional $40 billion and supplying 5 GW of TPU compute over five years. Amazon added $5 billion (up to $25 billion cumulative) along with a $100 billion, ten-year contract for Trainium compute. Microsoft and Nvidia are also reported to have joined with up to $5 billion and $10 billion respectively. Anthropic said on its official blog that annualized revenue (ARR) grew roughly 30x, from about $9 billion at the end of 2025 to $30 billion in April 2026.
• Why it matters: The old 'OpenAI–Google–Anthropic Big Three' framing has effectively collapsed. Google's core reason for putting $40 billion into a direct competitor is to lock in the largest buyer of its TPUs. It signals that the essence of the model race is moving from 'performance' to 'share of compute infrastructure'.

3. White House moves to reinstate Anthropic for federal agencies — access to the Mythos model under review

• Source: LLM Stats / international press roundup | 2026.04
• What happened: The White House is reportedly drafting guidance that would allow federal agencies to work with Anthropic again, and access to the new 'Claude Mythos' model is being reviewed alongside it. After Anthropic refused to let the US Department of Defense use its models for 'domestic mass surveillance and autonomous weapons', it had been excluded from some federal contracts; that could now be partially reversed.
• Why it matters: The conflict between AI ethics policy and access to the government market has entered a 'negotiation phase'. In Korea's public and defense sectors too, governance designs that review each model's use policy together with procurement eligibility look set to become essential.

4. Korea's industry ministry launches the combined 'M.AX·GX' call — 90 billion won across 9 programs

• Source: Ministry of Trade, Industry and Resources / AI Times | 2026.04.29
• What happened: The Ministry of Trade, Industry and Resources (Minister Kim Jung-kwan) officially opened the '2026 Smart Green Industrial Complex Support Program combined call' on April 29. A total of 90 billion won will go into nine programs covering manufacturing AI transformation (M.AX, Manufacturing AI Transformation) and carbon-free transition (GX) in industrial complexes. The combined call is meant to simplify applications and accelerate AI-driven process innovation at the industrial-complex level.
• Why it matters: It signals that the center of gravity in Korea's AI policy is shifting from 'conversational LLMs' to 'physical AI on the factory floor'. For mid-sized manufacturers and tenants of industrial complexes, it is a chance to fund computer vision, predictive maintenance, and digital twin projects.

5. Vercel publishes post-mortem on OAuth breach via an AI tool — supply chain risk in focus

• Source: Vercel Knowledge Base | 2026.04
• What happened: Cloud development platform Vercel published the follow-up findings on its April security incident. It began with a broad compromise of the Google Workspace OAuth app of 'Context.ai', a third-party AI tool an employee was using; the attacker used those permissions to take over the employee's personal account and then move into the internal environment. The same OAuth app is believed to have affected hundreds of other organizations. Vercel has since launched stronger environment variable management and a team-level security overview feature.
• Why it matters: As AI tools multiply, 'SaaS and AI apps holding OAuth permissions' have become a new supply chain attack entry point. Korean companies should now include OAuth scope reviews and environment variable governance in their standard procedures when adopting internal AI tools.

6. Stanford AI Index 2026 — "benchmark scores have surpassed human expert level"

• Source: IEEE Spectrum / Stanford HAI | 2026.04
• What happened: Stanford HAI has released its 2026 AI Index. On 'Humanity's Last Exam', where OpenAI o1 managed only 8.8% accuracy in 2025, scores rose to 38.3% within a year, and as of April 2026 Claude Opus 4.6 and Gemini 3.1 Pro have passed 50%. Global AI compute capacity has more than tripled every year since 2022, with Nvidia GPUs accounting for over 60% of global AI compute. Training xAI's Grok 4, however, is estimated to have produced more than 72,000 tons of carbon emissions.
• Why it matters: The model performance curve is still climbing steeply, and compute and energy consumption are exploding along with it. It is becoming clear that if Korea wants to be an AI powerhouse, it must design data center power supply and carbon policy together with the model race.

---

Today's insight

• AI power is moving from 'models' to 'compute infrastructure': Anthropic pulling in all four Big Tech giants as investors shows that the real axis of competition is now 'who can lock in more GPUs, TPUs, and Trainium chips'. The narrower the gap in model performance, the more decisive infrastructure share becomes.
• Courts and policy as the market's new variables: The Musk trial, the White House's review of reinstating Anthropic, and Korea's M.AX call are all cases of 'institutions', not 'technology', moving the market. Korean companies need to design a governance and procurement-eligibility roadmap alongside their technology roadmap.
• Field AI (physical and manufacturing AI) is the next category: As the Stanford index shows, model performance in chatbots and coding has entered its ceiling zone. 'AI that goes into the factory', like the industry ministry's M.AX call, is emerging as the area where Korea can do best.