Morning Digest — October 3, 2026
AI compute is moving into orbit, Starship has begun delivering its next generation of broadband satellites, and security teams now face vulnerability weaponization in less than a day. Today’s research radar also highlights practical advances in RAN slicing, passive LEO measurements, and agent-driven ISAC simulation.
Google puts AI compute into orbit with Project Suncatcher
Google has launched an experimental satellite carrying four Tensor Processing Units, giving Project Suncatcher its first in-space hardware test. The broader research program explores clusters of solar-powered satellites linked by free-space optics as future orbital AI data centers, with two Planet-built prototype satellites planned for early 2027.
The experiment matters beyond raw compute: radiation tolerance, thermal management, high-throughput optical links, and the economics of launch will determine whether orbital AI infrastructure can move from research into production.
Source: NPR
Starship delivers the first 26 Starlink V3 satellites
SpaceX confirmed that Starship’s first orbital mission deployed 26 Starlink V3 satellites. Each V3 spacecraft is designed to add roughly 1 Tbps of network capacity—about ten times that of earlier satellites launched on Falcon 9—and future Starship missions could carry substantially larger batches.
This is a pivotal link between SpaceX’s launch and communications businesses: operational Starship flights can accelerate constellation replenishment and expansion while adding capacity far faster per launch.
Source: SpaceX
Inseego completes its acquisition of Nokia’s FWA business
Inseego has completed its acquisition of Nokia’s fixed-wireless-access business, bringing indoor, outdoor, and millimeter-wave customer-premises equipment into its wireless broadband portfolio. The transaction expands Inseego’s reach across carrier markets in Europe, North America, Asia Pacific, and the Middle East.
Nokia receives about 1.9 million Inseego shares—roughly an 11% stake—and will provide $10 million to support engineering work connecting Inseego’s device software and cloud platform with parts of Nokia’s technology ecosystem.
Source: Nokia Newsroom
Microsoft says the cyber patching window has collapsed below 24 hours
Microsoft’s 2026 Digital Defense Report says AI-orchestrated malicious activity is now appearing in the wild. Its data puts the median interval between discovery of a vulnerability and weaponization at well under 24 hours, sharply narrowing the time defenders have to test and deploy patches.
The operational implication is clear: organizations increasingly need continuous asset discovery, identity hardening, automated prioritization, and compensating controls that can be deployed before a conventional patch cycle finishes.
Source: Microsoft 2026 Digital Defense Report
NVIDIA open-sources a policy-enforced runtime for autonomous agents
NVIDIA’s OpenShell provides a sandboxed runtime boundary for autonomous agents across open and closed models. It instruments the kernel to enforce policy over file access, system calls, and network connections, and formally checks proposed policy changes so teams can understand what new authority an agent would gain.
The project targets a central deployment problem: useful agents need broad tool access, but the same access expands the blast radius of compromised prompts, dependencies, or credentials. OpenShell makes those permissions explicit and independently enforceable.
Source: NVIDIA OpenShell on GitHub
U.S. charges a tech CEO over an alleged $300 million NVIDIA-chip smuggling scheme
Federal prosecutors charged California technology executive Greg Lui in an alleged scheme to export more than $300 million in controlled servers containing NVIDIA chips to China. Authorities say false paperwork obscured the final destination while shipments moved through Malaysia and Singapore.
The case shows export-control enforcement moving beyond manufacturers toward resellers, logistics routes, banking records, and documentation used to mask end users.
Source: Ars Technica
Qualcomm demonstrates agentic shopping through smart glasses
Qualcomm’s Snapdragon Summit demonstration with Mastercard showed an AI agent moving from personalized recommendations to a trusted transaction through smart glasses. Identity, authentication, payment security, privacy, and user control were treated as core pieces of the agentic-commerce workflow.
The demonstration points toward a near-term role for on-device AI: wearables can preserve local context and responsiveness while handing off authorization and settlement to established payment rails.
Source: Qualcomm
Research Radar
DRL-driven RAN Slicing Management: A V2X-oriented Approach In Multi-service Scenarios
Daniel E. Garcia-Fernandez and colleagues propose a deep-reinforcement-learning controller for coexisting safety-critical V2X and high-capacity eMBB slices. Validation on a real 5G standalone network shows fewer SLA violations than static and proportional allocation while preserving high resource utilization.
Source: arXiv
LEO Doppler Matching from Power Spectrum Data with Continuity-Based Segmentation and Multi-Position Clock Offset Estimation
Gaeun Kim, Seunghyeon Park, and Jongmin Park improve satellite association from passive SDR power-spectrum measurements. Across six hours of Starlink and OneWeb data, multi-position clock-offset estimation reduced receiver-localization error from 32.99 km to 1.84 km.
Source: arXiv
AIMS: An Agentic AI Framework for Sim-to-Real Multi-Modal ISAC
Yijie Bian and colleagues introduce a two-agent system that translates deployment requests into coordinated scene, sensing, wireless, and learning configurations. Experiments on DeepSense 6G report better vehicle detection and beam prediction, while structured knowledge and validation feedback improve planning correctness.
Source: arXiv
Source Notes
IEEE Xplore and ACM Digital Library searches returned no newly indexed target-topic papers, so today’s Research Radar uses verified arXiv records, including one paper submitted to ICCE-Asia 2026. Authenticated X access worked, though several rotated accounts had no substantive post inside the preferred 24–48-hour window.
Takeaway
AI infrastructure is spreading from devices to autonomous agents and orbit, while wireless research is making the connecting networks more adaptive, measurable, and dependable.