日本語
Sources: [1]
2026-07-15 Morning edition
AI News Daily
Morning edition

AI News Daily
2026-07-15

China becomes the first country to enforce a legal split between AI that works and AI that keeps you company, and the major platforms complied ahead of the deadline.

One story cleared the 48-hour freshness bar. A dated recap of the first half of July follows.
2026-07-15

Today's highlights

New today

2026-07-15
China's AI persona rules take effect. Doubao and Qwen shut off companion agents.Regulation and policy

Dated recap, first half of July

2026-06-30
Claude Sonnet 5 announced
2026-07-02
Anthropic passes OpenAI on revenue and valuation
2026-07-09
OpenAI ships the GPT-5.6 family: Sol, Terra, Luna
2026-07-13
Claude Code limits up 50%; Gemini 3.5 Pro due 17 July; EU labelling due in August
2026-07-14
Masayoshi Son projects 800 trillion yen a year of AI infrastructure by 2040
3
major Chinese platforms that withdrew persona agents on or before the deadline
Story 01

China's AI persona rules take effect

Published: 2026-07-15
Regulation and policy
Why it matters

This is the first regulation anywhere to draw a legal line between an AI that does work and an AI that offers companionship, and today it moved from drafting to live enforcement.

Source: Nikkei Asia — https://asia.nikkei.com/business/technology/artificial-intelligence/china-s-leading-chatbots-to-ditch-ai-personas-as-beijing-tightens-rules
Story 01

How the platforms responded

Published: 2026-07-15

ByteDance — Doubao

Ends its agent function today, on the day the measures take effect.

Alibaba — Qwen

Began a phased shutdown on 10 July and terminates the related services today.

Tencent — Yuanbao

Moved early and removed its persona-style agent function back in June.

Why it matters

Every major domestic platform complied on or ahead of schedule rather than contesting the rule. The precedent for anthropomorphic product design and for data-deletion policy is likely to travel beyond China.

Source: Bloomberg — https://www.bloomberg.com/news/articles/2026-07-06/bytedance-alibaba-pull-ai-companions-as-beijing-tightens-rules
Recap

The league table changed hands

Published: 2026-06-30 to 2026-07-13
  • Anthropic passes OpenAI (reported 2-3 July). About $30 billion annualised revenue and a $96.5 billion valuation, against $25 billion and $85.2 billion. First challenger to lead on both at once.
  • GPT-5.6 family (9 July). Flagship Sol, mid-tier Terra, fast Luna, released after a customer-by-customer review by the Department of Commerce.
  • Claude Sonnet 5 (30 June). 63.2% on SWE-Bench Pro at $2 input and $10 output per million tokens.
  • Gemini 3.5 Pro, due 17 July (reported 13 July). Architecture rebuilt from scratch, two-million-token window, Deep Think reasoning.
$96.5B
Anthropic valuation, against $85.2B for OpenAI
Dated recap of items reported in the first half of July. Source links for these items appear on the following two slides.
Recap

Two answers to the same capacity problem

Published: 2026-07-13 (Claude Code) and 2026-04-02 revision (Codex)

Claude Code — raise the ceiling

Weekly limits up 50%, extended to 19 July. Applied automatically to Pro, Max and Team, then reverting to standard.

Codex — meter past the ceiling

Token-based billing plus credits, revised on 2 April and settled into operation in July. Work continues on paid usage after the cap.

Why it matters

Usage allowance has become the competitive axis in developer tools. One vendor buys goodwill with headroom, the other sells continuity past the limit.

Sources: https://www.anthropic.com/news/higher-limits-spacex and https://developers.openai.com/codex/pricing
Recap

Capital, policy and labour

Published: 2026-07-03 to 2026-07-14
  • Masayoshi Son, 14 July. 800 trillion yen a year of AI infrastructure investment by 2040, keynoting SoftBank World 2026, with 100 trillion AI agents and one billion humanoids.
  • OpenAI, 3 July. Proposes granting the US government 1 to 5 percent of its equity at no cost, worth up to about $42.6 billion, as passive non-voting stock.
  • EU, 13 July. Transparency labelling for AI-generated content to be mandatory in August; a transparency code is under assessment as the mechanism.
  • US June employment data, 3 July. Of 142,000 tech job cuts in 2026, 88,000 are attributed to AI.
88,000
of 142,000 US tech job cuts in 2026 attributed to AI
Source: Nikkei (SoftBank World 2026 keynote) — https://www.nikkei.com/article/DGXZQOUC142BO0U6A710C2000000/
Trends

Trend overview

Race for the top spot

Anthropic against OpenAI, now contested on revenue and valuation as well as on models.

Usage allowance as a weapon

Developer tools compete on how much work a subscription actually buys.

Persona rules in force

Anthropomorphic-AI regulation moves from drafting to live enforcement.

The calendar ahead

2026-07-17
Gemini 3.5 Pro release
2026-07-19
Claude Code limits revert to standard
2026-07-31
Comment deadline on the FTC statement concerning AI accuracy
August 2026
EU transparency obligation begins
Wrap-up

Three things to remember

01
A regulator has, for the first time, legislated against a relationship rather than a capability, and the largest platforms complied on schedule.
02
The frontier league table changed hands in early July, and the contest has moved to supply and price rather than demonstrations.
03
Employment substitution is now visible in the statistics: 88,000 of 142,000 US tech job cuts this year are attributed to AI.
What to watch next

Gemini 3.5 Pro on 17 July, the Claude Code limit reset on 19 July, the FTC comment deadline on 31 July, and EU labelling in August. Both OpenAI and Anthropic filed confidentially for an IPO in May and June, making the listings from September the largest capital event of the second half.