China Desk — 15 Oct 2026

Twenty Chinese university students disciplined for using unauthorized VPNs

# China Digital Times

Harbin Engineering University warned 20 students for using unauthorized VPNs, triggering debate over academic access, competitive reporting and the return of informant culture.

The university says seven undergraduates and 13 graduate students were disciplined after a two-day enforcement period in late September.
Online discussion focused on an allegation that a student informed on a rival to improve graduate-school admission prospects; the supplied account reports this as rumor, not verified fact.
The episode exposes a structural conflict: researchers may need access to overseas academic sources while campus rules and the Great Firewall restrict that access.
Commentary archived by China Digital Times says reporting systems can be weaponized for personal disputes, producing self-censorship and mistrust.
The issue is not simply VPN compliance; it is how academic competition, constrained information access and disciplinary enforcement interact.

Open at China Digital Times Ask Claude ChatGPT

Shanghai AI Lab model helps prove a long-standing dynamical-systems conjecture

# Leiphone

Chinese coverage describes how the lab's Shusheng-S2 model supported a human-led proof of the three-dimensional Ghys conjecture, later checked using formal verification.

The researchers say the model produced a 400-page initial derivation, then supported seven rounds of follow-on reasoning as the problem graph expanded from 15 to 389 nodes.
The workflow paired long-horizon recursive reasoning with mathematicians who selected promising branches, rejected dead ends and redirected later exploration; the model did not independently deliver a finished proof.
The team says its training combined broad scientific pretraining with research-level mathematics data and verifiable-reward reinforcement learning, while a Harness system managed the extended reasoning process.
A separate team used the SHEAF framework to formalize and verify the proof, an important distinction between plausible mathematical prose and machine-checkable correctness.
The paper and formalization are reported as publicly available, allowing specialists to inspect both the argument and the verification artifacts.
If the result withstands peer scrutiny, it would be a notable demonstration of AI as a research collaborator; the broader claim that this is China's first world-class theorem solved with a domestic model is the outlet's framing.

Open at Leiphone Ask Claude ChatGPT

EU leaders consider a trade tool to eject Chinese firms from the market

# SCMP

Chinese companies are explicitly the target of a proposed EU trade instrument that could rapidly exclude them from the bloc, signaling a shift from tariffs toward market-access leverage.

The report says EU leaders broadly supported developing the tool, proposed in a French-German paper, and that the Commission president signaled work could begin.
A rapid-exclusion mechanism would make market access itself a strategic instrument, raising the stakes for Chinese firms with European operations and supply chains.
The proposal is not yet a law or an enforcement decision; its scope, trigger conditions and safeguards remain to be determined.
The article frames the initiative as an answer to geopolitical and economic coercion, but provides no examples of firms already facing the proposed mechanism.
For companies, the consequential question is whether the EU can define objective thresholds and coordinate enforcement across member states without creating broad uncertainty for ordinary trade.

Open at SCMP Ask Claude ChatGPT

China and US discuss an AI incident channel ahead of a leaders' summit

# SCMP

Chinese coverage leads with regular bilateral AI talks and plans for an incident hotline, while the immediate summit context is the effort to formalize a process before Xi-Trump meetings.

The reported goal is a basic framework for addressing AI incidents before the leaders meet at the APEC gathering in Shenzhen.
An incident channel could help manage risks such as agent failures, cyber operations or other fast-moving events, but the source gives no agreed protocol or scope.
A hotline is only useful if officials can exchange timely, actionable information and have clear procedures for escalation; those institutional details remain unsettled.
The effort sits alongside strategic competition, making it a narrow risk-management measure rather than evidence of broader trust.
The report does not establish that a final agreement has been reached.

Open at SCMP Ask Claude ChatGPT

Beijing and Washington pursue an AI incident framework before leaders meet

# SCMP

Chinese and US teams are discussing an AI incident framework ahead of a Xi-Trump meeting, making risk communication a potential deliverable in a relationship otherwise dominated by strategic competition.

The report says technical teams from both countries are working on a framework for handling AI-related incidents and institutionalizing bilateral dialogue.
The stated aim is to make AI cooperation a possible outcome of the leaders’ meeting on the sidelines of the APEC summit in Shenzhen.
A basic incident channel would be narrower than joint AI development: its immediate value would be reducing the chance that a technical failure or ambiguous event escalates into a political crisis.
The story presents the talks as an active diplomatic process, but does not specify which incidents would qualify, how information would be shared or whether either side has accepted binding obligations.
Any framework’s significance will depend on whether it creates durable points of contact and usable escalation procedures, rather than a summit announcement with no operational follow-through.
The report attributes its account to multiple sources, so the framework remains under negotiation rather than a completed agreement.

Open at SCMP Ask Claude ChatGPT

China opens a long-delayed bridge to deepen trade with North Korea

# SCMP

Premier Li Qiang used the opening of the New Yalu River Bridge to frame the link as a way to expand trade and people-to-people exchanges with North Korea.

The bridge opening is a concrete infrastructure signal after a long delay, though the report does not say when or at what capacity regular traffic will begin.
Beijing's public framing emphasizes friendship, trade and personnel exchanges, presenting connectivity as a tool for stabilizing and deepening the bilateral relationship.
The significance will depend on actual use: border policy, sanctions enforcement and North Korea's domestic economic conditions will shape whether the bridge carries meaningful trade.

Open at SCMP Ask Claude ChatGPT

Hong Kong suspends Chabaidao shares over suspected IPO demand manipulation

# TMTPost

Hong Kong regulators ordered Chabaidao trading suspended over suspected fabrication of IPO demand; Chinese coverage emphasizes alleged funding round-tripping and the risks to investors and franchisees.

The securities regulator says a suspected group may have helped create a false impression of demand for Chabaidao's 2024 offering; the investigation is ongoing, so the allegations are not established findings.
The report says the regulator restricted dealings in assets worth up to HK$4.379 billion (about $560 million) across accounts linked to the suspected group and six brokerages.
The alleged mechanism is not ordinary financial misreporting: the regulator suspects some IPO proceeds may have moved through undisclosed arrangements and returned to major share subscribers.
The article contrasts this with Luckin's accounting fraud, arguing that the suspected conduct concerns the offering process rather than the company's reported operating results.
Chabaidao's shares had already fallen sharply from their offer price, and the suspension leaves public investors unable to trade while the inquiry proceeds.
The consequences could extend beyond shareholders to the chain's more than 8,000 franchise outlets, though the report says daily operations had not yet been affected.
The case tests Hong Kong's willingness to intervene in an IPO after listing and highlights how weak demand and concentrated allocations can create incentives for questionable underwriting arrangements.

Open at TMTPost Ask Claude ChatGPT

Huawei details its chip-stacking roadmap and camera engineering

# AI Zhidx

Huawei used a two-day technical briefing to reveal more about its Kirin 9050 series, hybrid bonding and a multi-spectral camera system, signaling renewed emphasis on domestic engineering depth.

The report says Huawei has brought hybrid-bonding spacing to 1.5 micrometers and more than 50 million connections, though it does not provide independent chip measurements.
Its third-generation multi-spectral sensor is designed to improve color rendering across mixed lighting, skin tones and infrared interference.
Huawei also described a detachable 10x camera module and a broader effort to work around the physical limits of smartphone camera volume.
The event's significance is as much about communicating a long-horizon semiconductor roadmap as about any one handset feature; production yield and commercial performance remain unclear.
Huawei said it has run a 30-billion-parameter model on device, but the supplied report does not detail its speed, memory use or task quality.

Open at Zhidx Ask Claude ChatGPT

Ant Group's Ling-3.1-Flash targets long-running office-agent work

# AI QbitAI

Ant Group says its open Ling-3.1-Flash combines sparse experts with linear attention to handle long contexts and multi-step tasks at lower deployment cost.

Also: InfoQ China, Zhidx

The model reportedly has about 560 billion total parameters, 25 billion active per token and a context window of up to one million tokens.
Its attention stack uses seven KDA linear-attention layers for every Gated MLA layer, retaining periodic global attention while reducing sequence-processing cost.
The company says mixed-precision quantization cuts model-weight memory from roughly 534 GB to 292 GB and lowers deployment needs from eight Hopper GPUs to four in its test setup.
The model is aimed at office agents that must gather files, use tools, produce spreadsheets and presentations, and verify results rather than merely answer prompts.
Ant Group cites internal long-running coding and office evaluations, but the supplied material is promotional and does not provide independent reproduction.

Open at QbitAI Ask Claude ChatGPT

Top security researcher joins Chinese AI company Zhipu

# AI Zhidx

Security researcher Yuki Chen has joined Zhipu, and says he used GLM models to help find 23 vulnerabilities acknowledged by Microsoft, including a high-severity OpenSSH issue.

The report says the vulnerabilities passed Microsoft’s security-response review and were credited in its acknowledgments; the article does not provide technical write-ups for each one.
Zhipu says its models have helped identify more than 2,400 vulnerabilities in code repositories, with over 1,000 rated medium or high severity after expert review.
Cybersecurity is presented as an early commercial use for Chinese models, with about 100 security companies reportedly integrating GLM into products or workflows.
The strategic significance is dual-use: model-assisted vulnerability discovery can strengthen defense but also compress the time available to patch exposed systems.

Open at Zhidx Ask Claude ChatGPT

Qwen Office launches an agent tailored to e-commerce operations

# AI QbitAI

Alibaba’s Qwen Office introduced an e-commerce mode that connects merchant data and tools for research, content generation, store monitoring and ad optimization.

The product is designed to connect to existing seller tools and analyze authorized business data rather than require merchants to move all their systems.
Its functions cover trend research, product selling points, batch creative generation, multi-platform performance dashboards and return-on-investment alerts.
The report says a merchant used it to research a 3D-printed shoe shop and reduce launch preparation from about 30 days to seven, spending hundreds of yuan on tokens instead of several thousand yuan on services and production.
That example is an early case study, not a controlled assessment of sales lift or model accuracy.
The strategic play is an agent layer over fragmented commerce software: the product’s value depends on reliable connectors, permissions and domain-specific recommendations.

Open at QbitAI Ask Claude ChatGPT

StepFun open-sources a 600-billion-parameter model with 27 billion active

# AI Zhidx

Chinese lab StepFun released Step 5 Preview weights after the model rose to second place on OpenRouter usage rankings, emphasizing long-context efficiency and agent tasks.

The sparse MoE model has 600 billion total parameters, 27 billion active parameters and a claimed million-token context window.
Its architecture is 92 layers deep and relatively narrow, intended to preserve long-range information flow without the compute cost of a much wider model.
The model alternates gated sliding-window attention with sparse global attention; the team says it selects blocks of history before computing attention over a million-token context.
The company claims a benchmark score of 44 and lower task cost than Claude Opus 5, but those comparisons depend on the benchmarks and workloads selected.
Early usage on an open-model routing platform is a useful adoption signal, not proof of broad production performance.

Open at Zhidx Ask Claude ChatGPT

Kunlun’s open video model turns generated worlds into playable games

# AI Leiphone

Kunlun’s Matrix-Game 4.0 separates game-state management from video rendering and claims real-time play at 60 frames per second on one consumer GPU.

The system uses a lightweight game engine to maintain object state, rules and outcomes, while a video model renders the scene; an agent can generate the game logic.
The company reports 480p at 60 frames per second on an RTX 5090, under 150 milliseconds from input to response and a peak of 18.5 GiB of memory.
Explicit engine state addresses a weakness of video-only world models: visual continuity does not guarantee consistent health, inventory, collision or quest rules.
The architecture creates a reusable pipeline from code-generated interactions to structured visual constraints and generated frames, potentially lowering the asset burden for prototypes.
The project is open-sourced, but the announcement does not establish how long sessions remain coherent or how broad the game types are.
The key engineering question is whether the video renderer can stay synchronized with authoritative engine state as interactions become more complex.

Open at Leiphone Ask Claude ChatGPT

China's fifth plenum, EU trade tensions and Li Qiang's Russia trip

# Policy Sinocism

A China-focused briefing previews the fifth plenum and EU trade friction, notes Premier Li Qiang's planned Russia visit, and excerpts Xi Jinping warning against officials who publicly comply but quietly resist.

The briefing quotes Xi Jinping warning that officials can profess loyalty while obstructing central decisions in practice, a reminder that political discipline remains a recurring governance concern.
It flags preparations for the fifth plenum and a recent EU-China car trade arrangement, with Brussels signaling that the deal may not settle wider disputes.
It also lists Li Qiang's planned trip to Russia and changes involving foreign wind-power components as topics to watch.
The supplied excerpt is a roundup, not a full account of the plenum agenda or the outcomes of the diplomatic and industrial developments.

Open at Sinocism Ask Claude ChatGPT

Huawei explains how logic folding reshapes its Kirin 9050 chip

# AI Leiphone

Huawei says its Kirin 9050 Pro is the first mass-produced chip using its logic-folding approach, which places related circuits across layers to shorten critical signal paths.

The company reports more than 50 million hybrid bonds, over five million effective signal connections and a 1.5-micrometer bond pitch.
The design aims to improve timing and power by shortening routes between CPU cores and caches; the challenge depends on co-design across layout, interconnect and manufacturing.
Huawei reports gains in CPU, GPU and NPU performance and efficiency compared with its prior generation, but the supplied report does not offer independent benchmarks.
The separate technical briefing also describes fitting a 30-billion-parameter MoE model into a phone with under 2.5 GB of additional memory through tiered expert placement.
This is a meaningful domestic chip-design milestone if the performance claims hold at scale; the article does not identify the manufacturing process or independently verify yields.

Open at Leiphone Ask Claude ChatGPT

Apple says China-approved AI features will enter testing in late October

# Policy Zhidx

Chinese coverage stresses that Apple Intelligence and Alibaba models have approval but Siri AI is excluded; US coverage reports Cook's confirmation and the late-October test plan.

The initial test is said to include translation, writing assistance and photo cleanup, while a fuller release is expected later.
The Chinese report ties availability to local generative-AI filings and model partners, details absent from the US headline summary.
This is a China-specific rollout shaped by regulatory approval and domestic partners, not simply a delayed global software launch.

Open at Zhidx Ask Claude ChatGPT

Chinese interconnect-chip maker unveils a full-stack platform aimed at Broadcom

# AI Zhidx

Wuxi-based Zhongxing Micro says it has shipped more than 800,000 chips and is expanding from storage and networking into high-speed AI-cluster interconnects.

The company says its product range spans RAID and host-bus adapters, Ethernet and RDMA networking, PCIe switching and InfiniBand equipment.
It claims one RAID configuration beat Broadcom's write bandwidth by up to 47 percent and describes its PCIe switch as a substitute for a Broadcom part; the figures are company tests.
The strategic challenge is system qualification: interconnects must work across CPUs, operating systems, storage and communication libraries, not merely achieve a peak bandwidth number.
Partnerships with domestic server makers and operating-system vendors are intended to make the chips easier to deploy across Chinese infrastructure.

Open at Zhidx Ask Claude ChatGPT

EU hardens its China trade stance after concluding pressure can work

# Policy SCMP

The EU-China hybrid-car deal has not softened Brussels’ approach: the bloc is pursuing dialogue while preparing tougher action on Chinese trade practices, according to SCMP.

The report says European officials believe the recent hybrid-car arrangement shows that threats can produce results, rather than requiring a broader retreat from trade enforcement.
Brussels is pursuing a two-track approach: continue intensive dialogue with Beijing while also strengthening measures to counter practices it considers unfair.
That framing suggests the EU sees negotiation and trade defense as complementary tools, not mutually exclusive choices.
The article does not specify the next enforcement measures or demonstrate that member states agree on every element of the tougher line.
For Chinese firms in Europe, the practical signal is that a negotiated settlement in one sector may not prevent scrutiny in others.

Open at SCMP Ask Claude ChatGPT

Apple's China AI launch may use both Alibaba and Baidu services

# Analysis QbitAI

Chinese coverage emphasizes Qwen's role in Apple's approved initial release, while the US roundup focuses on the test date; separate reporting also points to Baidu services in search and Siri-related features.

Also: Leiphone

The article says Apple's China-facing AI stack may combine on-device and cloud models with local services, rather than using a single provider for every capability.
The reported integration of Baidu search and visual search is based partly on media reporting and a third-party code inspection, not an Apple product announcement.
Apple's first approved features are described as a limited subset of its global AI offering; the newer Siri experience is not expected in the initial release.
This is a reminder that market entry for AI features in China can depend on multiple local model and service partners as well as regulatory approval.

Open at QbitAI Ask Claude ChatGPT

StepFun opens weights for a long-horizon agent model

# AI InfoQ China

StepFun released BF16 and FP8 weights for Step 5 Preview, a 600-billion-parameter sparse model designed for long-context software and knowledge-work agents.

Also: IT Home

The model activates 27 billion parameters per token and supports a one-million-token context; sparse activation reduces compute per token but does not remove the need to manage the full model weights.
Its 92-layer design alternates gated sliding-window attention with sparse global attention, while Sparse GQA selects blocks of past tokens to reduce long-context attention work.
The team also trains agents to compress task state when context fills, treating memory management as part of the learned policy rather than relying only on a large context window.
The article cautions that lower indexing overhead does not mean end-to-end inference becomes eight times faster; communication, weight access and tool execution remain costs.
Open weights let companies test cross-file edits and recovery in their own environments, but the source provides no independent task-completion comparison.

Domestic chipmakers rush to support StepFun's new open model IT Home

Open at InfoQ China Ask Claude ChatGPT

Alibaba's new AI glasses trade a display for lighter weight and always-on sensing

# AI InfoQ China

Chinese coverage links the launch to a reported 8% domestic smart-glasses shipment decline, while describing Alibaba's response: lighter, screen-free glasses with a low-power sensing path.

The N1 Pro is reported at about 44 grams and uses a 5-milliwatt always-on sensing path that watches for basic scene changes before waking higher-power components.
The design avoids running a full video model continuously; a low-power chip monitors simplified grayscale frames, then escalates to communications, audio or imaging functions when needed.
Alibaba says persistent sensing requires user authorization and does not default to uploading the captured scene, addressing the privacy concern inherent in wearable cameras.
The trade-off is responsiveness: the reviewer reports one- to two-second delays for some voice and cloud interactions, with performance dependent on network access.
Chinese market data cited by the article show the category weakening, suggesting comfort and a convincing daily use case may matter more than adding sensors or translation features.

Open at InfoQ China Ask Claude ChatGPT

Chinese graduates face opaque AI-writing scores in thesis reviews

# Culture TMTPost

Graduates report being called back or delayed over AI-detection scores that vary across tools, raising questions about whether universities can fairly use an unexplained probability as an academic integrity test.

The report describes students facing new AI-writing checks even after their theses had passed review, with some asked to provide drafts, interview records and other evidence of their own work.
One student's paper reportedly received a 15% score from one tool and 39% from another, illustrating how results can depend on the detector rather than a consistent standard.
Unlike plagiarism matching, AI detectors often provide a percentage without identifying a source or showing which reasoning led to the score.
Some students say they rewrote formal academic prose to reduce the number, even when the edits made the writing less precise or natural.
University rules also differ on what AI assistance is permitted, so a detector score alone cannot establish whether a student used AI for research, editing or authorship.
The dispute exposes a governance gap: institutions are adopting measurable thresholds before they can explain error rates, appeal procedures or the meaning of a flagged score.

Open at TMTPost Ask Claude ChatGPT

Embryo researcher returns from the US to lead a lab in China

# Culture SCMP

Stem-cell scientist Wu Jun says China has become a leading center for embryo research, returning to Westlake University after working in the United States.

Wu’s research includes stem-cell-derived embryo models, a field that can reveal early development but also raises ethical and regulatory questions.
His move is a talent-flow signal, but the article offers his assessment rather than comparative staffing or publication data.
The story matters because China’s research capacity can attract specialists even as scientific collaboration and regulation remain politically sensitive.

Open at SCMP Ask Claude ChatGPT

Former Huawei AI lead starts a company focused on proactive physical AI

# Analysis QbitAI

Former Huawei assistant-model lead Tang Duyu has founded Guanxing Shuzhi to build systems that infer when people need help and act through physical devices without waiting for a direct command.

The proposal changes the agent problem from executing a stated task to interpreting a continuous stream of events and deciding what to do, when to intervene and which device should act.
That adds a difficult safety layer: the same observed behavior can signal an emergency, a harmless choice or an ambiguous situation where intervention would be intrusive.
The founder's background spans assistant models, code intelligence and large-scale device deployment, but the report gives no product demonstration or company funding details.
Proactive behavior could make connected devices more useful, but it also raises questions of consent, liability and who sets acceptable intervention boundaries.

Open at QbitAI Ask Claude ChatGPT

Young Chinese viewers increasingly question harmful filial expectations

# Culture SCMP

A report on online reactions to family reality television describes younger Chinese viewers rejecting coercive versions of filial duty and arguing that family relationships need boundaries.

The article uses public reactions to television stories about parent-child relationships to show how viewers debate responsibility, emotional harm and what children owe parents.
Its central contrast is between traditional filial expectations and a newer emphasis on mutual respect and healthier boundaries.
Reality television gives these disputes a shared public stage, where viewers examine family behavior and openly judge power dynamics that were once treated as private matters.
The item offers a cultural reading rather than polling or evidence that attitudes have shifted uniformly across generations.

Open at SCMP Ask Claude ChatGPT

Chinese AI infrastructure engineers see sharp wage increases

# Culture Leiphone

Recruiters and executives say pay for strong AI infrastructure engineers has risen sharply as labs compete to improve GPU utilization and reduce inference costs.

The article describes reported salary growth from about 300,000 yuan (roughly $42,000) in 2024 to around 1 million yuan (about $140,000) in 2026 for one candidate; exceptional packages can be much higher.
The work includes kernel optimization, inference scheduling and parallelism strategies such as expert parallelism and prefill-decode separation, where small utilization gains can save large compute budgets.
One source estimates typical GPU utilization at 25-27% and says strong teams can push it toward 40%; these are industry estimates, not an audited survey.
The change reflects a shift in AI competition from model quality alone toward the systems engineering that determines how much useful output each accelerator produces.
The article's salary anecdotes should not be read as market-wide averages, especially where equity makes up a large share of startup compensation.

Open at Leiphone Ask Claude ChatGPT

SeePhys Hub collects synchronized vision, touch and whole-body motion data

# AI Leiphone

Chinese startup Poxiao and Tianlong Mobile launched a wearable system intended to capture full-body human demonstrations, including tactile contact, for robot learning.

The system combines a four-camera headband, wrist devices, body trackers and tactile gloves to record scene context, hand pose, pressure and whole-body movement together.
Its premise is that first-person video alone misses how people shift posture and apply force, information needed to train mobile manipulation rather than isolated hand actions.
The company says its software segments and labels recordings, reconstructs scenes and retargets human movements to robot bodies.
The key research question is data efficiency: whether richer multimodal demonstrations improve learning enough to justify greater capture complexity and hardware burden.
The announced system is an engineering product, not evidence that a robot trained on its data generalizes across homes, workplaces or different robot designs.

Open at Leiphone Ask Claude ChatGPT

China buys half of Brazil's oil exports as Hormuz disruption shifts trade

# Signal SCMP

Brazil-China trade reached a reported $143 billion in the first nine months of the year as China absorbed a large share of Brazilian oil exports during disruption to Gulf supply.

Brazilian exports to China rose 12% to $84.7 billion, while imports from China grew 9.9% to $58.6 billion, according to the Brazil-China Business Council.
The figures show how an energy shock can redirect China's commodity sourcing and deepen bilateral trade well beyond the oil sector.
A large increase in Brazilian crude does not by itself show that China can replace Gulf suppliers at equivalent scale or quality.
The source attributes the trade shift to the Iran war disrupting Hormuz; the supplied excerpt does not break out oil volumes or contract duration.

Open at SCMP Ask Claude ChatGPT

Chinese apps block ByteDance's screen-operating phone assistant

# Policy 36Kr

A Chinese roundup reports that 151 apps, including WeChat, Taobao and Meituan, opted out of ByteDance's GUI automation protocol, limiting the assistant's ability to act across apps.

The apps remain installable and usable by people; the refusal targets automated screen reading and simulated taps.
The dispute is about who controls the interface when an AI agent operates a phone: app developers can deny automation even when the device maker offers an agent.
The source says the opt-out list includes major social, commerce and productivity applications, creating a substantial obstacle to cross-app task completion.
The same roundup includes unrelated US corporate and market news; the Chinese app-access conflict is the distinctively China-relevant item.

Open at 36Kr Ask Claude ChatGPT

Chinese researchers use flight data to locate suspected electronic-warfare jamming

# Analysis SCMP

Chinese military researchers say they used civil aviation data to estimate the locations of suspected GPS-spoofing sources near Israel, illustrating how public datasets can support military sensing.

The researchers reportedly analyzed civil aviation flight data to identify interference sources affecting aircraft navigation during the Iran conflict.
They estimated two suspected units near Haifa and reported positioning errors of roughly 11 kilometers and 15 kilometers.
The result is a claim by the research team, not independent confirmation that the sources were Israeli military units; the article says the researchers did not explain why they focused on Israel.
The approach suggests a low-cost form of open-source geolocation: interference visible in civilian navigation records can reveal information about emitters otherwise difficult to observe.
The reported accuracy may be useful for broad-area situational awareness, but it is not precise enough on its own to identify a specific platform or establish attribution.
The dual-use significance lies in converting routine commercial data into a military intelligence input, a capability relevant to electronic-warfare competition.

Open at SCMP Ask Claude ChatGPT

As AI makes code cheaper, Chinese engineering leaders say complexity rises

# Analysis InfoQ China

Chinese engineering leaders argue that AI-generated code raises the bar for architecture, context and testing: the bottleneck is shifting from writing software to keeping larger, faster-changing systems correct.

Interviewees say code generation can sharply increase output per developer, but also expands the volume of changes a person must understand and own.
They emphasize project context and engineering harnesses: a model needs business rules, repository history and architecture knowledge, not just a prompt and a codebase.
One leader expects testing to move into the coding loop, especially where agents cannot observe real device behavior or end-to-end outcomes.
The discussion reflects a management tension: conventional productivity metrics may become less informative when a small team can produce far more code.
These are practitioner views collected ahead of a conference, not measured evidence that development teams have already achieved the cited productivity gains.

Open at InfoQ China Ask Claude ChatGPT

Chinese drugmaker puts its information department under the president

# Analysis 36Kr

Simcere’s president has taken direct control of the information department and is embedding AI staff in business teams, treating organizational adoption as a prerequisite for drug-discovery gains.

The company says it is using AI in early target discovery and molecule design, while deploying technical staff into departments to move from experimentation toward routine use.
Simcere’s innovation-drug revenue now accounts for a reported 85% of first-half revenue, creating pressure to find more differentiated early-stage assets.
Its leadership argues that future licensing deals need globally relevant, original targets rather than only mature projects first designed for the domestic market.
The approach ties AI adoption to a strategic change in Chinese pharma: improving discovery quality and global competitiveness, not simply automating office tasks.
The article reports management plans and company figures, not evidence that AI has yet improved clinical success rates.

Open at 36Kr Ask Claude ChatGPT

Apple crosses $5 trillion as China AI rollout draws closer

# Analysis QbitAI

Chinese coverage pairs Apple’s $5 trillion market milestone with the long-delayed China rollout of Apple Intelligence, stressing that local users may get the older feature set without the new Siri.

The article says Apple closed above $5 trillion in market value while broader technology shares fell, but the China-relevant development is the expected local AI launch.
The reported first wave includes writing, notification and photo tools, with Alibaba models and possible Baidu services supporting localization.
The China rollout is described as a regulatory and partner integration problem; the article does not provide a confirmed launch date or final feature list.

Open at QbitAI Ask Claude ChatGPT

Hunan police warn rural men about cross-border marriage scams

# Culture SCMP

Police in Hunan warned that paid international matchmaking can expose rural men to financial loss and legal trouble, highlighting the risks around a tightly regulated marriage-broker market.

The police post warned that some cross-border arrangements lack genuine emotional foundations and may involve illegal agencies charging high fees.
For-profit international matchmaking is illegal in China, according to the report.
The warning frames marriage as a consumer-protection and public-order issue, while the underlying concern also reflects demographic and gender imbalances that can shape demand for such services.
The report does not provide data on prevalence or prosecutions.

Open at SCMP Ask Claude ChatGPT

China's biggest apps just decided an AI agent has to be invited in

151 apps told ByteDance's phone agent no

Beijing waves through Apple's easy features

Exports surged and Brussels hardened

Zhipu's cyber bet, and idle GPUs

Harbin's informer, Luckin's algorithm

Threads we are pulling