2026.05.11 AI News Briefing — Pentagon signs AI deals with 8 companies, leaving out only Anthropic; KOSPI breaks 7,000

The new week opened with the global AI power map dividing one step further. The US Department of Defense signed AI tool contracts with eight Big Tech firms including SpaceX and OpenAI at once, officially excluding Anthropic, and the list included dark-horse newcomer 'Reflection AI', founded by an AlphaGo co-developer. Meanwhile, Anthropic signed a compute token deal with SpaceX and launched AI agents for financial work. Colorado is on the verge of passing a bill that repeals and replaces its own AI Act, and in Korea the KOSPI crossed 7,000 for the first time ever on AI and semiconductor tailwinds.

---

1. Pentagon signs AI deals with 8 companies including SpaceX and OpenAI — Anthropic excluded, dark horse is AlphaGo-alumni Reflection

• Source: CNN Business / SiliconANGLE / Breaking Defense / Al Jazeera | 2026.05.01–10
• What happened: The US Department of Defense simultaneously signed contracts with eight companies, SpaceX, OpenAI, Google, Microsoft, Nvidia, AWS, Oracle, and Reflection AI, to use AI tools on its classified networks (Impact Level 6 and 7 environments). The military described it as 'an accelerating step toward transforming the US armed forces into an AI-first fighting organization', and also stressed the principle of avoiding 'vendor lock'. Its own 'GenAI.mil' platform is already used by 1.3 million personnel, with hundreds of thousands of AI agents built on it. Anthropic, by contrast, was left off the list after refusing the Pentagon's demand that it be allowed to use Claude for 'any lawful purpose', including autonomous weapons and mass surveillance, and then receiving a 'supply chain risk' designation. The White House recently reopened some talks after Anthropic's new model announcement, but the company did not make it into this round of eight contracts.
• The most noteworthy company on this list is not a Big Tech firm but Reflection AI. It is a young lab co-founded in March 2024 by Ioannis Antonoglou, a co-developer of AlphaGo, and Misha Laskin, former Gemini reward modeling lead at Google DeepMind; in October 2025 it raised a $2 billion Series B led by Nvidia at an $8 billion valuation. Its core assets are the autonomous coding agent 'Asimov' (released July 2025) and an in-house LLM trained on tens of trillions of tokens, and it positions itself as 'the Western alternative to DeepSeek (an open-source frontier lab)'. It is especially unusual that a company with no officially launched commercial product landed on a 'classified-network grade (IL6/7)' contract list.
• Why it matters: Exactly ten years after AlphaGo beat Lee Sedol 9-dan in Korea, a company founded by one of AlphaGo's creators has entered the Pentagon's classified networks. It is a strong signal that the 'dark horse variable' in the AI power map no longer lives only inside Big Tech. At the same time, the trend of an AI model's 'use policy' deciding access to the government market hardened another notch. Korea's defense, public, and financial sectors should now treat 'use policy fit' and 'backend developer governance', beyond 'performance', as effectively mandatory steps in model selection.

2. Anthropic–SpaceX compute deal revealed — "tokens have begun to eat the economy"

• Source: Semafor / Futureseek Daily Link | 2026.05.10
• What happened: Analysis reported that Anthropic and SpaceX have signed a large compute token deal. In the same period, SpaceX holds a $60 billion option to acquire AI coding IDE Cursor. Between late April and early May, Anthropic secured an additional $40 billion from Google, $5 billion from Amazon (plus a $100 billion AWS compute package), and chip deals worth hundreds of billions of dollars with Google and Broadcom, and its annualized revenue (ARR) of about $30 billion is tallied as having overtaken OpenAI (about $24 billion).
• Why it matters: The trend of models, cloud, space, robotics, and coding IDEs being bound into a single 'token economy' is starting to affect the Korean market too. Korean companies need a multi-layer strategy that escapes single-vendor dependence and ties together 'token cost, agent governance, and power and chip supply'.

3. Anthropic unveils AI agents for financial work — automating pitchbooks and credit memos

• Source: CNN Business | 2026.05.05
• What happened: Anthropic unveiled new AI agents for financial work. They are designed to carry multi-step tasks that junior to mid-level analysts used to perform, such as writing pitchbooks and credit memos, all the way to completion. This extends the same company's earlier statement that 'the job of software engineer itself will disappear and become builder'. A Microsoft report based on a survey of 20,000 workers in 10 countries likewise found that most companies are at the stage of redefining 'which skills within a role humans still need' rather than changing the role itself.
• Why it matters: The data reconfirms that AI is at the stage of 'redefining jobs', not 'taking jobs'. In Korea's finance, legal, and consulting fields too, a workflow where 'the agent drafts the output of junior to mid-level analysts and humans review and decide' is likely to become the standard.

4. Colorado's SB 189, repealing and replacing its AI Act, nears passage — US chatbot and health AI legislation accelerates

• Source: Troutman Privacy + Cyber + AI / US state legislature reports | 2026.05.04–11
• What happened: Colorado's SB 189 (which repeals and replaces the existing 'Colorado AI Act') has passed both chambers, with only Senate concurrence on the House amendments remaining. The same state also passed a bill banning 'surveillance pricing and wage setting' (HB 1210), and two bills on chatbots and healthcare AI are likely to pass before the session ends on May 13. Iowa's chatbot law (SF 2417) took effect with the governor's signature, Hawaii passed SB 3001, and the New York Assembly passed the 'AI Training Data Transparency Act' (A 6578). Vermont is pushing mental health AI regulation (HB 816) and a 'neurological rights' bill (HB 814) at the same time.
• Why it matters: US state-level AI legislation is spreading along four axes: chatbots, health, employment, and training data. For Korean AI and SaaS companies entering the US market, a 'state-by-state compliance matrix' has become effectively mandatory, and it will directly affect the certification requirements for the 'task-completing agents' that Maum AI and AIWORKX emphasized on the closing day of AI EXPO KOREA.

5. KOSPI breaks 7,000 for the first time ever — "Korea and Taiwan lead the AI rally after the Iran war ends"

• Source: Newspim (citing Bloomberg) / Click Today | 2026.05.10–11
• What happened: Korea's KOSPI crossed 7,000 for the first time in history and is taking another run at settling above 7,500. In an analysis dated the 9th, Bloomberg reported that 'the Korean and Taiwanese markets have risen the most in the world this month, with Samsung Electronics, SK hynix, and TSMC emerging as the core beneficiaries of the AI boom'. The KOSPI is up 78% for the year, and analysts argue the memory chip industry has moved from 'cyclical' to 'earnings-driven structural growth'. Investment is spreading to the materials sector as well: Doosan's electronics business group is expanding investment in high-spec CCL for AI accelerators 2.8-fold, and Lotte Energy Materials is converting its Iksan plant to circuit foil and HVLP copper foil lines.
• Why it matters: The trend Microsoft's 'Global AI Diffusion Report' identified, Korea as 'the fastest-rising country', has been confirmed at the same time in capital market indicators. For the next six months, the keyword for Korean industry analysis is likely to be the synchronized 'AI–semiconductor–materials–power' cycle.

6. CNN: "AI isn't taking jobs, it's rewriting them"

• Source: CNN Business / Google, Stack Overflow, and Microsoft surveys | 2026.05.10
• What happened: CNN published a roundup analysis arguing that 'AI does not take a person's job wholesale; it automates some steps within the job'. Per a Google survey, 90% of tech workers use AI at work, and 84% of Stack Overflow respondents use or plan to adopt AI tools. Boris Cherny, who leads Claude Code, said that 'by year's end the title software engineer itself will disappear, replaced by something like builder'. In its report based on a survey of 20,000 workers in 10 countries, Microsoft diagnosed that 'companies have not yet redesigned employee evaluation metrics and incentives'.
• Why it matters: This should be read together with the layoffs at Snap and Oracle and JPMorgan's reclassification of AI as infrastructure. In Korean HR, work to reflect 'AI usage intensity' in evaluations and incentives is expected to begin in earnest soon, and it sums up the key macro message of the month of May.

---

Today's insight

• AI power is being reorganized around 'policy + tokens + alumni networks', not 'technology': The Pentagon's eight contracts and Anthropic's exclusion, the Anthropic–SpaceX compute token deal, and the surprise inclusion of AlphaGo-alumni Reflection — the common message of three events in one week is that 'use policy, compute access, and talent networks, more than model performance, divide the market'. Korean companies' AI strategy now has to be designed along three axes at once: technology roadmap, use policy, and token and power supply.
• Korea is at the stage where 'AI gains' show up in both capital markets and industry: The KOSPI crossing 7,000, Microsoft naming Korea the top AI adoption riser, the industry ministry's 52.75 billion won AI Factory program, Upstage's 560 billion won raise, and the 'physical, action-oriented AI' demos at AI EXPO KOREA all happened within a single month. The key variable for Korean IR and industry analysis in the coming quarters is the synchronized 'AI–semiconductor–materials–power' cycle.
• The 'job replacement' narrative is quickly shifting to 'role redesign': Anthropic's financial agents, the Claude Code lead's 'builder' remark, and Microsoft's 20,000-worker survey all point to the same message: 'AI does not cut people, it changes the makeup of their work'. The keyword for Korean HR and education over the next six months is likely to be 'role redesign' rather than 'reskilling'.