Morning Digest — November 5, 2026
Today’s strongest signals converge on networks and autonomous systems: Starlink Mobile is building a terrestrial 5G layer, Nokia expects AI to dominate traffic growth, and AI agents are moving deeper into vulnerability research, biology, coding, and robot evaluation.
SpaceX taps Qualcomm for Starlink Mobile’s nationwide 5G network
SpaceX selected Qualcomm’s Dragonwing Cellular Infrastructure platforms for a nationwide U.S. terrestrial 5G network under Starlink Mobile. The multi-year rollout is scheduled to begin in 2027 and will span tens of thousands of cell sites.
The architectural significance is the integration of satellite and terrestrial access under one operator. Qualcomm’s entry also introduces a major U.S. silicon supplier into a RAN market long dominated by Ericsson, Nokia, and Samsung.
Source: Qualcomm
Nokia says AI will generate 44% of network traffic by 2035
Nokia’s 2026 Global Network Traffic Report forecasts total monthly traffic will nearly quadruple from 2026 levels to 3,789 exabytes by 2035. AI-related workloads are projected to account for 44% of that total.
The composition matters as much as the volume: Nokia expects agentic-AI traffic to grow 160-fold and physical-AI traffic 105-fold. Networks therefore need to support distributed inference, machine-to-machine interaction, and latency-sensitive control rather than simply carry more human-generated video.
Source: Nokia
OpenAI’s autonomous pipeline finds 948 Linux privilege-escalation flaws
OpenAI says an autonomous vulnerability-discovery, exploitation, and patching pipeline found 948 unique local privilege-escalation vulnerabilities in the Linux kernel during one research run. The system validated findings and generated candidate patches, not merely speculative bug reports.
This is a striking demonstration of cyber automation at industrial scale. It strengthens the case for AI-assisted defensive auditing while also showing why controlled access, coordinated disclosure, and model safeguards are becoming urgent infrastructure concerns.
Source: OpenAI
Claude helps map the first 3D structure of a bacterial vault particle
Anthropic’s biology team reports the first 3D structure of a vault particle found in prokaryotic cells. Vaults are large barrel-shaped protein assemblies whose biological function remains unclear, and related structures had previously been associated with complex eukaryotic cells.
Claude accelerated parts of the structural analysis from a workflow that could take weeks or months to one completed in days. Anthropic also documents simple model mistakes and the expert intervention needed to correct them, offering a useful picture of AI as a scientific collaborator rather than an autonomous oracle.
Source: Anthropic
Meta acquires multi-agent coding startup Conductor
Meta acquired Conductor, a startup whose product lets software engineers launch and manage multiple coding agents in parallel workspaces. Financial terms were not disclosed.
The deal is another sign that coding assistants are evolving into orchestration systems. The competitive unit is increasingly a managed team of agents working across tasks and repositories, not a single chatbot producing code snippets.
Source: TechCrunch
RoboColosseum opens real-world evaluation across 100+ robots
RoboColosseum launched as an open initiative for evaluating robot models on more than 100 physical robots and over 140 in-the-wild scenes. The Hugging Face-backed program is offering 10,000 free evaluation hours and publishing rollout data, including failures, on the Hub.
Real-world robot evaluation is expensive and hard to reproduce. A shared physical benchmark could make embodied-model comparisons more credible than curated demonstrations and expose failure modes that simulation misses.
Source: Hugging Face
Research Radar
StarWorld: A Cross-Layer World Model for Starlink Prediction and Streaming
Bowen Zhang, Pengcheng Luo, Hongjun Xie, Shujin Jia, Genke Yang, and Jian Chu introduce a world model that connects stochastic Starlink network dynamics with executable throughput prediction, adaptive bitrate control, and playback processes. A calibrated trust gate determines when imagined futures are reliable enough to use.
On real Starlink measurements, StarWorld achieved the lowest future-distribution errors at the principal horizons and reduced pairwise controller quality-of-experience effect error by 76%.
Source: arXiv
Energy-efficient End-to-End Communication System with Spiking Neural Networks
Pooja Plavilayil Anil, Te-Chuan Chiu, and Nguyen Van Huynh propose a spiking-neural-network encoder for end-to-end communications. Sparse spike activity, temporal modulation, and weighted aggregation are used to reduce inference energy on constrained devices.
Across five antenna configurations, the encoder maintained comparable bit-error performance at suitable timesteps while reducing estimated encoder energy by 68.93%, 52.20%, and 35.69% at two, three, and four timesteps respectively.
Source: arXiv
6G Uplink Waveform and Rank Selection under Distributed Reception
Yewen Cao, Fen Hou, and Yulin Shao study joint selection among CP-OFDM and DFT-s-OFDM at rank 1 and noncoherent rank 2 under distributed reception. They derive waveform-switching thresholds and a rank-selection margin based on transmit power, despreading loss, and spatial multiplexing gain.
The analysis shows that access-point deployments with similar antenna budgets and aggregate channel gains can still favor different waveforms and ranks because spatial and frequency structure determine the best mode.
Source: arXiv
Source notes
Searches of IEEE Xplore and the ACM Digital Library returned no newly indexed matching papers. The research section therefore uses verified arXiv records published within the last seven days.
Takeaway
Satellite-terrestrial convergence and AI-native traffic are reshaping network architecture just as autonomous agents move from software generation into security, science, and physical-world evaluation.