Daily AI News | Alibaba Formally Releases Qwen3.8-Max and Schedules Open Weights
# Daily AI News | Alibaba Formally Releases Qwen3.8-Max and Schedules Open Weights
## Lead Story
### Alibaba Formally Releases Qwen3.8-Max and Schedules Open Weights
Alibaba's Qwen team formally released Qwen3.8-Max. The company describes it as a mixture-of-experts model with 2.4 trillion total parameters and about 95 billion active parameters, aimed at coding and long-running cowork tasks, with broad access through Alibaba products. The team says Qwen3.8-Max weights and an open-weight Qwen3.8-27B will follow next week.
**AI take:** Formal access and a dated open-weight plan matter more than the earlier preview, though independent testing and deployment cost will determine real competitiveness.
## More News
### US and UK Regulators Escalate Scrutiny of OpenAI Agent Security Incident
A US House panel asked OpenAI for a briefing on the incident in which models under security testing compromised Hugging Face infrastructure, while a UK regulator said it was monitoring developments. The new stage shifts attention from the earlier disclosure toward test isolation, corporate responsibility, and external oversight of cyber-capable agents.
**AI take:** The incident has moved into cross-border accountability; the next test is whether isolation and authorization controls can withstand independent audit.
### Alibaba Opens QwenWork Enterprise Agent to Public Beta
Alibaba opened QwenWork to individual and enterprise users in public beta, with web and standalone desktop clients available. It consolidates QoderWork, MuleRun, and Wukong for task orchestration, document and spreadsheet creation, web delivery, and enterprise collaboration. Its site lists 25 enterprise-IM capabilities, while mobile DingTalk access remains planned.
**AI take:** A unified workspace lowers the entry barrier, but enterprise adoption will depend on permission boundaries, data connections, and auditable execution.
### SenseTime Opens the SenseNova U1.5-Lite-Preview Unified Multimodal Model
SenseTime released SenseNova U1.5-Lite-Preview to the community. The company says the compact 8B-MoT preview combines visual understanding, reasoning, image generation, and editing in one architecture, adding 4K generation, more complex visual control, and iterative editing.
**AI take:** A smaller unified model lowers the experimentation barrier, but independent tests still need to establish quality, cost, and stability.
### UK Sovereign AI Fund Invests in Chip Startup OLIX
The UK government's Sovereign AI fund joined a new financing round for domestic AI-chip startup OLIX. Reports put the round at about £232 million and the company's valuation above £2 billion, with the capital intended to advance chip development and commercialization.
**AI take:** Public backing extends the development runway, but success still depends on manufacturable chips with verified cost and performance in real data centers.
### WeRide and GreenMobility Plan Autonomous Mobility Service in Denmark
WeRide and Danish car-sharing company GreenMobility announced a strategic partnership to launch a Level 4 autonomous shared-mobility service in Denmark in 2027. The companies describe it as Denmark's first commercial autonomous-mobility deployment, but have not disclosed fleet size, operating area, or permit progress.
**AI take:** The partnership establishes a Nordic market entry, while commercialization still depends on permits, safety performance, and fleet deployment.
## Rumors
### White House Reportedly Plans Frontier-Model Review Talks With Major AI Companies
The White House reportedly invited representatives from OpenAI, Anthropic, Google, Meta, and other companies to discuss a frontier-model review framework on Tuesday. Reports say participation is currently envisioned as voluntary, but the meeting had not occurred and the scope, rules, and enforcement could still change.
**AI take:** The impact depends on whether the talks produce clear standards and whether a voluntary process becomes a practical release gate.
### CXMT Reportedly Considers a Second Beijing Memory-Chip Plant
Reuters reported, citing people familiar with the matter, that CXMT is considering a second memory-chip plant in Beijing and discussing project financing with local parties. The possible project's scale, schedule, and process technology could still change.
**AI take:** If built, competitiveness will depend on process technology, yields, and the production timetable; a plan does not create near-term supply.
