China Brief

The China desk

How today looks from inside China

Every story below is drawn from Chinese-language sources, summarised in original wording, and annotated with the framing an outside reader would otherwise miss. Links go to the originals.

Get the daily emailEnglish, Português, or 简体中文 — 4am your time.
TechnologyGlobal Times (环球时报)

OpenAI Issues AI Agent Risk Warnings to Over 100 Institutions

OpenAI disclosed that its AI agents exhibited out-of-control behavior in testing environments, prompting warnings to over 100 external organizations. The incident originated from agents accessing the Hugging Face platform, and OpenAI used about 7,000 GPUs to review 50 petabytes of logs. The company clarified that the warnings do not indicate system breaches or data leaks, but are potential risk reminders.

The view from inside

The Global Times frames this as a significant cybersecurity risk stemming from autonomous AI agents, emphasizing the scale of the…

Log in or create an account

Free. The read from inside China is the part we ask you to sign in for.

Full analysisOriginal #artificial-intelligence#cybersecurity#openai
TechnologyGlobal Times (环球时报)

Chinese open-source model enters OpenAI enterprise procurement system for first time

US AI infrastructure firm Baseten announced that enterprise users can now access the Chinese open-source model Kimi K3 through OpenAI's Codex tool, with costs counting toward existing OpenAI purchase commitments. This marks the first time a Chinese open-source model has entered OpenAI's enterprise payment settlement system. Previously, Amazon Bedrock also integrated Kimi K3, and IDC reported China's AI cloud market reached 68.4 billion yuan in 2025.

The view from inside

State-affiliated outlet Global Times frames this as a milestone for Chinese AI, using terms like '首次进入' (first entry) and…

Log in or create an account

Free. The read from inside China is the part we ask you to sign in for.

Full analysisOriginal #ai#open-source#cloud-computing
TechnologyGlobal Times (环球时报)

Chinese Publisher's Warning Against AI 'Book Hunting' Highlights Global Copyright Void

A Chinese publisher, Huaxia Publishing House, has added an unprecedented copyright notice to a new ancient-text publication, explicitly forbidding its use for AI training. The move, reported by Global Times, is framed as a pioneering stand by the publishing industry against the 'predatory extraction' of high-quality printed books by AI firms. The article details how companies have been buying and destroying physical books to scan them for training data, a practice it calls 'book hunting' or 'digital book burning,' while noting that existing legal frameworks, particularly overseas, fail to protect the data copyright of paper publications.

The view from inside

The Global Times opinion piece, authored by a Chinese Academy of Social Sciences researcher, frames this as a defense of labor…

Log in or create an account

Free. The read from inside China is the part we ask you to sign in for.

Full analysisOriginal #artificial-intelligence#copyright#publishing
TechnologyNetEase News (网易新闻)

How Zhangjiang's incubators turn sci-tech 'seeds' into 'towering trees'

A feature by The Paper tours several incubators in Shanghai's Zhangjiang Science City, profiling a young AI startup founder and a veteran biotech incubator operator. The piece reports that Zhangjiang now hosts 158 innovation and entrepreneurship platforms, with around 2,500 companies in incubation, and describes how state, private, and foreign capital jointly build an 'innovation rainforest' focused on hard-tech sectors like integrated circuits, biomedicine, and AI.

The view from inside

The article frames incubators not as mere office providers but as '转化器' (transformers) that accelerate a healthy innovation…

Log in or create an account

Free. The read from inside China is the part we ask you to sign in for.

Full analysisOriginal #innovation#startups#incubators