A U.S. federal jury in San Francisco concluded an 11-day trial by convicting former Google engineer Ding Linwei on seven counts of economic espionage and seven counts of theft of trade secrets. Prosecutors allege that between May 2022 and April 2023, he stole and uploaded over 2,000 pages of confidential documents related to Google's AI supercomputing architecture, TPU/GPU clusters, AI workload scheduling software, and SmartNIC networking technologies, while concealing his collaborations with two Chinese tech companies and ties to a startup plan in Shanghai. Sentencing is pending.

According to Reuters, China has approved domestic AI startup DeepSeek's purchase of NVIDIA's H200 AI chips, subject to regulatory conditions still being finalized. The move indicates that while supporting leading model firms in expanding compute capacity, authorities remain cautious about approving critical chip imports and their intended uses. If conditions are clarified and deliveries proceed smoothly, DeepSeek's compute constraints in large model training and inference are expected to ease, potentially influencing its model iteration pace and service offerings.

Reuters reports that SpaceX has applied to the U.S. FCC for approval to build solar-powered 'satellite data centers' to support artificial intelligence computing. This concept extends data center infrastructure into space, aiming to address power supply and deployment location constraints, but still faces regulatory and engineering challenges including spectrum and orbital resource allocation, communication links, and compliance. The application reflects how growing AI compute demand is driving innovation in data center form factors and energy sources.

Based on euaiact.com's analysis of Article 72 of the EU AI Act, providers of high-risk AI systems must establish and document a 'post-market monitoring' system proportionate to the technical characteristics and risks of their systems. This includes continuously collecting, recording, and analyzing data from deployers or other sources to assess ongoing compliance with Chapter III, Section 2 requirements throughout the system’s lifecycle, including analysis of interactions with other AI systems when appropriate. The European Commission must adopt an implementing act by February 2, 2026, providing a template and list of elements for such monitoring plans, which will be incorporated into technical documentation.

At an 'AI+' event in Xiong'an New Area, the 'Jishu' general-purpose structured data large model (LimiX) was launched, jointly developed by Tsinghua University and Wenzhen Intelligence (Xiong'an). Positioned for general modeling and analysis of structured data, the model reportedly supports around 10 types of tasks including classification, regression, high-dimensional representation extraction, and causal inference, integrating causal reasoning and synthetic data generation to enhance generalization and interpretability. It has been validated in over 100 scenarios across more than 20 industries, targeting real-economy applications in manufacturing, energy, power, and petrochemicals.

Kling AI announced the launch of its next-generation Kling 3.0 series multimodal models and opened early global beta testing, including 'Kling Video 3.0', 'Kling Video 3.0 Omni', and 'Kling Image 3.0'. Public information indicates the series follows an 'All-in-one' product philosophy, covering image generation, video generation, video editing, and post-processing workflows, extending capabilities along film production and content creation pipelines. As the models remain in beta, details on access scope, pricing, performance metrics, and available API formats have not yet been disclosed.

Google has launched the 'Gemini for Students' program, offering eligible enrolled students one year of free subscription to Google AI Pro. Benefits include access to Gemini 3 Pro, NotebookLM, 2TB of cloud storage, and various creative features such as image generation and Deep Research. Applicants must be at least 18 years old, enrolled at a qualifying institution in designated regions, and verified via SheerID. Subscriptions will auto-renew at $19.99/month if not canceled before expiration. The free enrollment window is marked to close on January 31, 2026.

Mandiant disclosed attack activities linked to ShinyHunters: attackers used targeted voice phishing (vishing) and spoofed corporate-branded phishing sites to trick employees into submitting SSO credentials and MFA codes, which were then relayed in real time to complete authentication. After gaining access, attackers registered their own MFA devices to maintain persistent access. Once inside, they leveraged Okta, Microsoft Entra, or Google SSO consoles to move laterally across SaaS platforms such as Salesforce, Microsoft 365, Slack, and Dropbox, exfiltrating data in bulk. Mandiant also released detection rules and hardening recommendations.

The Ministry of Public Security (MPS) is soliciting public opinions on the draft 'Cybercrime Prevention Law (for comments)', a seven-chapter, 68-article proposal emphasizing source-level governance of cybercrime and regulation of underground/gray-market industrial chains. Key provisions include strengthening real-name registration and management of suspicious accounts, banning interference with real-name verification; defining prevention obligations for online payment, traffic promotion, and other services; advancing a trusted digital identity strategy and national network identity authentication public service; and establishing cross-border cybercrime prevention mechanisms, including international sanctions, service oversight, and travel restrictions. Feedback is due by March 2, 2026.

Intel released version 1.3 of LLM-Scaler-vLLM, offering a Docker-based deployment solution to run the vLLM inference and serving framework on Intel Arc Battlemage GPUs. This update primarily expands the range of large language models compatible with the platform, improving deployment flexibility for developers and enterprises seeking local inference and building low-latency, high-throughput model services on Intel GPUs. No specific performance data, throughput/latency benchmarks, or comparisons with other GPU platforms have been disclosed in available materials.