AI Daily Briefing

Thursday, July 02, 2026

Thursday, July 02, 2026 (Asia/Seoul)

To: kimkj008@gmail.com

1. Anthropic unveiled Claude Sonnet 5 and made it the default model for Free and Pro. In Claude Code, it also became the default model with a 1 million token context window, and an introductory price of $2 input and $10 output per 1 million tokens will apply through August 31.

2. On June 30, the U.S. Department of Commerce lifted export controls on Claude Fable 5 and Mythos 5, and Fable 5 reopened globally on July 1. Anthropic added a new cybersecurity classifier up front that blocks more than 99% of jailbreak attempts.

3. The Japanese government decided to invest 1 trillion yen over five years, about KRW 8.6 trillion, into domestic foundation models, and through the SoftBank-led Noetra consortium it has set its direction toward physical AI and robotics.

===== 6 Categories =====

Anthropic unveils Claude Sonnet 5, shifts to default model for Free and Pro (2026-07-01)

Sonnet 5 is a model focused on handling browsers and terminals on its own and pushing through multi-step work to completion. It has become the default model in Free and Pro, and it is also used in Max, Team, and Enterprise. Claude Code now carries a 1 million token context window. The introductory price is $2 input and $10 output per 1 million tokens, rising after August 31 to $3 input and $15 output.

The Federal Trade Commission released a draft policy statement addressing conduct that suppresses the accuracy of AI answers. The vote was 2 to 0, and comments will be accepted through July 31. The draft views the Colorado AI law as pressuring businesses to alter model outputs, and says it is implicitly preempted to the extent it conflicts with federal regulation. It cites President Trump's executive order from last December as the basis. The policy statement, however, is not a formal rulemaking, so it cannot by itself invalidate state law, and the final judgment belongs to the courts.

Japan announces sovereign AI models and a plan for 10 million AI robots (2026-07-01)

The Japanese government released domestic foundation model plans together with a revised AI robotics strategy. Development will be led by the SoftBank-led Noetra consortium, and it places physical AI at the front, trained on multi-sensor data, robot teleoperation records, and CAD-grade simulation. The initial model is aimed at limited deployment by the end of 2028.

The export controls imposed on June 12 were lifted on June 30. The trigger was a circumvention method reported by Amazon researchers, including a case in which Fable 5 identified software vulnerabilities and generated exploit code. Commerce's CAISI tested both the old safeguards and the new safeguards before agreeing to lift the controls. Mythos 5, which had also been blocked for key infrastructure institutions inside the United States, was partially restored as well.

Fable 5 reopened on the Claude platform, Claude.AI, and Claude Code. Pro, Max, Team, and some Enterprise users can use Fable 5 for up to 50% of their weekly usage limit through July 7. The new classifier filters risky cyber-task requests, and instead of rejecting blocked requests, it routes them to Claude Opus 4.8 and notifies the user. The tradeoff is false positives. Ordinary coding and debugging requests may also be flagged more often than before and routed to Opus.

Japan to invest 1 trillion yen in domestic AI over five years, Noetra consortium to expand to 44 companies (2026-07-01)

The scale of government support is up to 1 trillion yen over five years, about KRW 8.6 trillion. Noetra, which includes SoftBank and Sony, is expected to add automotive, electronics, finance, and logistics companies, bringing participating firms to 44. It is the execution plan built on the long-term 2040 investment framework announced at the end of June.

===== Review of U.S. AI Developments =====

Technology: Anthropic unveiled Claude Sonnet 5. It focuses on completing multi-step work and self-checking, and in Claude Code it became the default model with a 1 million token context window.

Law / Regulation: The FTC released a draft AI accuracy policy statement, vote 2 to 0, comments through July 31, asserting preemption against the Colorado AI law. The Department of Commerce lifted export controls on Fable 5 and Mythos 5.

Product / Use: Fable 5 reopened, including in the United States, and a new classifier was attached that routes blocked requests to Opus 4.8. Sonnet 5 became the default model in Claude Code.

===== Review of China AI Developments =====

Technology / Product: [Follow-up] DeepSeek will release the official V4 version in mid-July and divide pricing by time of day. Beijing time 9 a.m. to 12 p.m. and 2 p.m. to 6 p.m. are peak periods, priced at double the off-peak rate. For V4 Pro at peak pricing, cache-miss input is 6 yuan per 1 million tokens and output is 12 yuan. By moving away from fixed pricing toward a system in which prices change with demand, like electricity and cloud services, this is being read as a sign that China's AI market is moving from price-cut competition to competition over compute efficiency.

Investment / Business: If DeepSeek's time-based pricing model takes hold, other Chinese model providers are likely to follow the same approach.

===== Today's AI Use Tip =====

Using Sonnet 5's 1 million token context window for long-document review

1) In Claude Code or the Claude platform, confirm that the default model is Sonnet 5. In Free and Pro, it is already the default.

2) Paste in contracts, court rulings, or extensive meeting transcripts at once and have it extract the list of issues first. Even if you chain several steps, it tends to carry the work through instead of stopping midway.

3) After receiving the draft, ask it to point back to the supporting statutory provisions and the cited locations so it can check its own work.

4) If you see a notice that a coding or security question was routed to Opus 4.8, it may be a false positive from Fable 5's new classifier, so rewrite the question in a clearer business-work context.

5) The introductory pricing lasts through August 31. If you batch high-volume review work into this period, you can avoid the higher pricing that begins in September.

===== Attorney Kim Kyung-jin Commentary =====

===== [Source Links] =====

Anthropic Claude Sonnet 5 official: https://www.anthropic.com/news/claude-sonnet-5

Sonnet 5 benchmark and pricing comparison (MarkTechPost): https://www.marktechpost.com/2026/06/30/anthropic-claude-sonnet-5-vs-sonnet-4-6-vs-opus-4-8-agentic-coding-benchmarks-api-pricing-and-cost-performance-tradeoffs-compared/

Fable 5 reopening and export-control lift (CNBC): https://www.cnbc.com/2026/06/30/anthropic-says-trump-admin-has-lifted-export-controls-on-claude-fable-5-and-mythos-5.html

Where Fable 5 was globally restored (VentureBeat): https://venturebeat.com/technology/anthropic-is-bringing-back-claude-fable-5-globally-after-us-lifts-export-control-order-where-can-enterprises-access-it

How Fable 5's new classifier works (MarkTechPost): https://www.marktechpost.com/2026/07/01/anthropic-redeploys-claude-fable-5-on-july-1-after-us-export-controls-lift-adds-new-cybersecurity-classifier/

FTC AI accuracy policy statement (FTC): https://www.ftc.gov/news-events/news/press-releases/2026/07/ftc-seeks-public-comment-policy-statement-addressing-ai-accuracy

Japan sovereign AI and robotics plan (Japan Times): https://www.japantimes.co.jp/news/2026/07/01/japan/japan-ai-plans/

Japan $6.2 billion AI initiative (Digitimes): https://www.digitimes.com/news/a20260701VL202/japan-ai-investment-industrial-infrastructure.html

DeepSeek V4 time-based pricing (Digitimes): https://www.digitimes.com/news/a20260701PD207/deepseek-llm-api-price-gpu.html

DeepSeek V4 mid-July launch (TechNode): https://technode.com/2026/06/30/deepseek-to-launch-v4-in-mid-july-with-new-peak-time-api-pricing/

Chrome extension Grabbit tip: You can drag across the links above with the right mouse button to open several at once in new tabs, or hold the Z key and drag with the left mouse button to copy only the link addresses.

===== Search Keywords =====

United States: "AI news July 1 2026 OpenAI Anthropic Google", "Claude Sonnet 5 launch July 2026", "Anthropic Fable 5 restored export controls lifted July 2026", "FTC AI accuracy policy statement Colorado preemption July 2026", "AI copyright lawsuit ruling July 2026", "AI funding chip July 1 2026"

Japan and others: "Japan AI news July 1 2026 SoftBank Noetra physical AI", "WAIC 2026 世界人工智能大会 7月"

Generated at: Thursday, July 02, 2026 06:00 KST

The core issue is not how hard it is, but whether it solves the problem. That is planning in the end. The real moat is not code, but experience, trust, and data.