China's Ministry of Industry and Information Technology issued joint guidelines Monday with three sister agencies to overhaul the country's internet basic resources, calling for major advances by 2030 and a more advanced national infrastructure by 2035, TechNode reported. The document directs research into agent-to-agent networks, satellite internet, digital identity infrastructure and IPv6 upgrades, and calls for integrating AI, blockchain and distributed identifiers into the resource system. The new target builds on work covered in AIPD's June 26th edition, when China's State Administration for Market Regulation released seven national standards for AI agentAI agentAn AI system that carries out multi-step tasks on its own, such as browsing, writing code or making purchases, rather than answering a single prompt. Agents raise new questions about liability, security and oversight because they act rather than just advise. interconnection covering identity, discovery and tool invocation.
Read at TechNode ↗
Ant Group's AI Safety Lab has open sourced two guardrail models, SingGuard-NSFA for autonomous agentsAI agentAn AI system that carries out multi-step tasks on its own, such as browsing, writing code or making purchases, rather than answering a single prompt. Agents raise new questions about liability, security and oversight because they act rather than just advise. and SingGuard for multimodal systems, TechNode reported. SingGuard-NSFA is designed to detect prompt injection, sensitive data theft, malicious code execution, resource abuse and permission misuse before agents take action, and comes in four sizes from 0.8 billion to 9 billion parameters. Ant said the model can make a single risk judgment in about 50 milliseconds and covers seven major risk categories across 133 languages. The release lands five days after China's industry ministry issued a "backdoor" alert on Anthropic's Claude Code.
Read at TechNode ↗
Zhipu AI on Monday completed a placement of new H shares raising approximately HK$31.4 billion (about $4 billion), per Jiemian. The company priced 19.78 million new shares at HK$1,588 (about $200) apiece, roughly a 13% discount to the prior close. The raise deepens a summer of megafundraises among Chinese frontier labs; MiniMax's HK$16 billion (about $2 billion) round appeared in AIPD's July 10th edition. In a separate internal letter reported by LatePost, founder Tang Jie said Zhipu will launch a "Touch High" plan focused on artificial general intelligenceAGIArtificial general intelligence: an AI system that can do most economically valuable cognitive work at or above human level. There is no agreed test for it, which is why debates about when it will arrive, and what to do about it, are so contentious. (AGI) research rather than short-term commercialization, prioritizing long horizon tasks, autonomous agentAI agentAn AI system that carries out multi-step tasks on its own, such as browsing, writing code or making purchases, rather than answering a single prompt. Agents raise new questions about liability, security and oversight because they act rather than just advise. systems and mechanistic interpretability.
Read at Jiemian ↗
Chinese researchers have built an all-optical interconnect linking standard electronic chips that they claim boosts AI distributed inferenceInferenceRunning a trained AI model to answer a query, as distinct from training it. Inference is where most ongoing compute and energy demand now sits, and it is what chip export rules increasingly target alongside training. speeds by more than 100 times while using just one-ninth of the typical computational resources, the South China Morning Post reported. Corresponding authors from Peking University published the work in the journal National Science Review. The comparison covers distributed inference on conventional GPU stacks rather than a single accelerator benchmark, and independent evaluations of the reported numbers have not been published outside the paper. Chinese researchers have increasingly pitched optical and photonic architectures as an alternative to piling on more GPUs, a route constrained for Chinese labs by U.S. controls on Nvidia's frontier accelerators.
Read at South China Morning Post ↗
As Typhoon Bavi bore down on eastern China, state broadcaster China Media Group warned that hobbyist weather bloggers posting AI generated typhoon forecasts on social media could run afoul of the law, the South China Morning Post reported. Some had drawn on open source AIOpen-weight modelAn AI model whose trained parameters are published so anyone can download, run and modify it, as opposed to one reachable only through the developer's API. The main policy tension is that open weights spread capability widely and cannot be recalled once released. weather models to predict landfall zones for specific locations, and a few were charging fees for tailored outputs. Under China's Meteorology Law, only official meteorological stations may issue public weather forecasts and warnings through a centralized release system. The rebuke arrives as Chinese authorities have already moved to bring AI generated content under labeling and traceability rules, extending existing content controls to a class of open weight scientific models that had largely escaped the framework.
Read at South China Morning Post ↗