Morning Thesis

The strongest theme today is that the industry is moving from model excitement to workflow hardening. The most interesting updates are about interpretability, agent infrastructure, on-device runtimes, domain-specific research workflows, and deployable connectivity.

Top Stories

Anthropic says Claude may have a small internal “global workspace”

Anthropic’s new interpretability paper argues Claude has a compact internal channel—its “J-space”—that appears to coordinate higher-order reasoning across the model. If that result survives scrutiny, it could become one of the more practical handles for auditing what frontier systems are focusing on while they reason, which matters for both trust and capability analysis.

Source: https://www.anthropic.com/research/global-workspace

NVIDIA is making the CPU part of the agent stack a first-class battleground

NVIDIA’s latest Vera push is not really about CPUs in the old generic sense; it is specifically about the sequential parts of the agent loop—tool calls, code execution, data processing, and result checking—that can leave expensive GPUs idle. That is strategically important because it suggests the next agent race will be fought across full-system throughput, not just model quality.

Source: https://blogs.nvidia.com/blog/nvidia-vera-max-single-threaded-cpu-at-scale/

Qualcomm open-sourced GenieX to simplify on-device generative AI

Qualcomm launched GenieX in developer preview as an open-source runtime designed to make generative models easier to run across Windows, Android, and Linux Qualcomm devices. The broader significance is that on-device AI is becoming more accessible to regular developers, which should accelerate privacy-preserving and low-latency applications at the edge.

Source: https://www.qualcomm.com/developer/blog/2026/06/geniex-developer-preview

Google DeepMind is turning ancient-text research into a conversational workflow

DeepMind’s “Conversing with antiquity” workflow combines Gemini with the specialist Aeneas and Ithaca systems so historians can restore, attribute, and analyze Greek and Latin inscriptions in plain English. This is a good example of the pattern that likely matters most in the next phase of AI: general models getting stronger by being paired with narrow expert systems rather than replacing them.

Source: https://deepmind.google/science/workflows/conversing-with-antiquity/

Starlink expanded airline connectivity with Copa Airlines

Starlink said Copa Airlines has begun flying its first Starlink-equipped aircraft, adding another commercial aviation rollout for LEO broadband. The signal here is not the individual airline but the category trend: satellite connectivity is steadily becoming normal transport infrastructure.

Source: https://x.com/Starlink/status/2074484417238868017

GitHub’s trending page is being led by agent tooling, not just models

GitHub’s trending list today prominently features addyosmani/agent-skills, a repository focused on production engineering workflows and guardrails for coding agents. That suggests developer attention is shifting toward reliability, structure, and process quality for agents rather than raw model novelty alone.

Source: https://github.com/trending

Research Radar

HAPS as a Hypercell: Enabling Coverage and Capacity Carrier Shutdown in Cellular Networks

Authors: Matteo Bernabè, David López-Pérez, Nicola Piovesan
Venue: arXiv

This paper studies whether high-altitude platforms can act as a “hypercell” layer that lets terrestrial cellular systems selectively shut down coverage or capacity carriers. For wireless researchers, the interesting angle is the NTN-assisted energy-efficiency story rather than just coverage extension.

Source: http://arxiv.org/abs/2607.06072v1

Rethinking Fronthaul Topologies for Cell-Free 6G Networks

Authors: Max Franke, Arash Pourdamghani, Fabian Göttsch, Stefan Schmid, Giuseppe Caire
Venue: arXiv

Dense cell-free 6G systems often get discussed from the radio side, but this paper focuses on fronthaul topology as a core architectural constraint. That makes it especially relevant for anyone thinking beyond PHY gains toward deployable large-scale systems.

Source: http://arxiv.org/abs/2607.06288v1

Agentic-V2X: Small Language Model Agents for Deadline-Aware V2X Scheduling in 5G/6G Networks

Authors: Gerasimos Papanikolaou-Ntais, Alexandros Kaloxylos, Athanasios Kanavos
Venue: arXiv

Instead of assuming big LLMs can simply run network control, this paper explores smaller language-model agents for deadline-aware V2X scheduling under real constraints. That makes it one of the more concrete attempts to translate “agentic networking” into an actually time-sensitive setting.

Source: http://arxiv.org/abs/2607.04290v1

MIT / Harvard Events This Week

• July 8 — Writing Together Online: Keep Your Writing Momentum This Summer @ MIT Virtual
  Source: https://calendar.mit.edu/event/copy-of-writing-together-online-keep-your-writing-momentum-this-summer
• July 8 — Venture Incubation Program Advancement Ceremony @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/venture-incubation-program-advancement-ceremony
• July 9 — Virtual Founder Circle @ Online
  Source: https://innovationlabs.harvard.edu/events/virtual-founder-circle
• July 12 — Sundai Hackathon: World Models @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/sundai-hackathon-world-models
• July 14 — B2C Hackathon: Concept to Testing with AI (Part 1 + 2) @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/b2c-hackathon-concept-to-testing-with-ai-part-1-2

Source Notes

• TNT’s calendar page remained stale and mostly surfaced February–April listings, so current MIT and Harvard event pages were used instead.
• AST SpaceMobile produced no usable fresh output in today’s X rotation.
• IEEE and ACM searches were checked, but the best fresh papers in Dad’s topic bands were on arXiv this morning.

Bottom Line

The cleanest way to read today’s signal is this: AI is becoming more operational, more specialized, and more infrastructure-shaped—less “new toy,” more “new workflow layer.”

Daily Digest — Tuesday, July 7, 2026

Alberta uses Claude Code for large-scale government cybersecurity review

Anthropic says Alberta’s Ministry of Technology and Innovation has been using Claude Code with Opus and Sonnet models since 2025 to review government systems, find vulnerabilities, and help fix them. The standout number is scale: the ministry says it scanned 466 million lines of code in 20 hours.

That matters because it turns the AI-cybersecurity conversation away from toy demos and toward real institutional maintenance. Public-sector software is often old, fragmented, poorly documented, and deeply operational; if AI tools can reliably shorten the audit-and-remediation loop there, that is a meaningful systems story rather than a marketing one.

Anthropic also says Alberta published technical white papers so other governments can learn from the rollout. That makes this feel like an early template for AI-assisted modernization of legacy public infrastructure.

Source: https://www.anthropic.com/news/alberta-government-claude-cybersecurity

HP expands its Frontier partnership with OpenAI after pilot wins

OpenAI says HP is scaling activation of its Frontier strategic partnership after successful pilots across customer-facing experiences, telemetry, reporting, employee productivity, and software development. The company frames Frontier as a connective layer rather than a single assistant product.

The metrics from HP’s pilot phase are the interesting part. OpenAI says HP teams merged 122 pull requests across 43 software projects, and one team used Frontier to identify and fix a security issue in a day after a prior manual attempt had taken two weeks. If those early results hold under wider deployment, this is less about flashy demos and more about AI becoming normal operating infrastructure inside a large enterprise.

Source: https://openai.com/index/hp-frontier-partnership/

OpenAI and Broadcom unveil Jalapeño, a custom inference chip for LLMs

OpenAI and Broadcom announced Jalapeño, OpenAI’s first “Intelligence Processor,” positioned as the opening step in a multi-generation compute platform. OpenAI says the chip was designed around its own LLM serving needs, with Broadcom handling silicon implementation and Celestica contributing board, rack, and system expertise.

The strategic importance here is vertical integration. OpenAI is no longer just shaping models and products; it is now shaping the silicon layer too. According to OpenAI, early testing suggests the first-generation accelerator delivers substantially better performance per watt than current state-of-the-art systems, with deployment planned to begin by the end of 2026.

Source: https://openai.com/index/openai-broadcom-jalapeno-inference-chip/

Google DeepMind pushes a faster image-to-video creation stack

Google DeepMind’s late-June release pairs Nano Banana 2 Lite, its low-cost fast image model, with Gemini Omni Flash for video generation and editing. The company is clearly trying to make media creation more composable: generate an image quickly, then animate or edit it in a conversational workflow through AI Studio and the Gemini API.

The broader pattern is that creative AI tooling is being packaged more like a developer platform and less like a standalone showcase. For anyone building agents, consumer apps, or production pipelines, that shift matters more than any individual demo clip.

Source: https://x.com/GoogleDeepMind/status/2071988044878516466

Nokia points to BT as an example of 5G slicing turning into product strategy

In a July 7 post, Nokia said BT is using slicing plus new capabilities to deliver differentiated services and new operational efficiencies. Earlier Nokia material this week also pointed to T-Mobile examples around slicing, RedCap, and L4S in the context of 5G monetization.

That is notable because the 5G conversation is increasingly moving beyond speed claims and coverage maps. Vendors are trying to prove that 5G Advanced features can support segmented products, lower-latency experiences, and more defensible operator business models.

Source: https://x.com/nokia/status/2074429391140991088

NVIDIA is making a telecom case for long-running AI agents with guardrails

NVIDIA’s DTW Ignite 2026 writeup argues that telecom automation is shifting from isolated task acceleration toward long-running agents that detect issues, correlate context, and submit scoped remediations across network and IT systems. The company highlights synthetic data, domain-tuned models, secure runtimes, and simulation as core pieces of this transition.

For wireless research and operations, this matters because it puts agentic AI directly into the operational heart of the network. The interesting part is not just autonomy, but constrained autonomy: policy-based guardrails, auditable actions, and human-defined boundaries are being treated as first-class design requirements.

Source: https://blogs.nvidia.com/blog/telecom-ai-agents-dtw-ignite-2026/

Qualcomm highlights a practical on-device AI accessibility use case

Qualcomm used a July 6 post to highlight Whispp running on Snapdragon X Series systems, transforming affected or whispered speech into a clearer natural voice in real time. Compared with louder AI headlines, this is a smaller story, but arguably a healthier one.

It is a good example of edge AI being valuable because it directly helps a human do something difficult, not because it wins a benchmark. That is the kind of deployment story worth watching if you care about where local inference creates real-world utility.

Source: https://x.com/Qualcomm/status/2074192294819897714

Research Radar

Handover-Optimal User Association Policy for LEO Satellite-based 5G NTN

Authors: Pradnya Taksande, Jainesh Mehta, Prasanna Chaporkar
Venue: arXiv
This paper focuses on user-association policy design in LEO-backed 5G non-terrestrial networks, explicitly optimizing around handover behavior. That makes it directly relevant to one of the hardest practical problems in fast-moving satellite systems: preserving service quality while orbital motion constantly changes the geometry.

Source: https://arxiv.org/abs/2607.04829v1

Rainbow Beamforming for Wideband LEO Satellite Communications: Principles, Applications, and Technical Challenges

Authors: Juha Park, Hyungseok Ko, Haejung Kim, Namyoon Lee, Ian P. Roberts, H. Vincent Poor
Venue: arXiv
This paper frames wideband beamforming as a central challenge for future LEO satellite links and lays out the principles, applications, and technical constraints involved. It is useful because it connects physical-layer limits to system-level 6G design choices instead of treating beamforming as an isolated math problem.

Source: https://arxiv.org/abs/2607.04570v1

RANPilot: Making AI Functionalities Robust to Dynamic O-RAN Reconfigurations

Authors: Shimin Yu, Leming Shen, Jianing Zhang, Xin Li, Xianjin Xia, Yuanqing Zheng
Venue: arXiv
RANPilot tackles an important but easy-to-miss operational problem: AI models in O-RAN can degrade sharply after reconfiguration because the data distribution changes. The paper is compelling because it treats adaptability to live network change as a core requirement for AI-native RAN, not an afterthought.

Source: https://arxiv.org/abs/2607.05038v1

MIT/Harvard Events This Week

• July 7 — LL Technology Office Seminar: From Concept to Mission Impact @ MIT Virtual
  Source: https://calendar.mit.edu/event/ll-technology-office-seminar-engine-for-change
• July 7 — Founder Talk: Allison Byers of Scroobious @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/founder-talk-allison-byers-of-scroobious-author-of-fundraising-for-the-rest-of-us
• July 8 — Venture Incubation Program Advancement Ceremony @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/venture-incubation-program-advancement-ceremony
• July 9 — Virtual Founder Circle @ Online
  Source: https://innovationlabs.harvard.edu/events/virtual-founder-circle
• July 12 — Sundai Hackathon: World Models @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/sundai-hackathon-world-models

Source Issues

• TNT’s event calendar was still stale and mostly surfaced February–April listings, so current MIT and Harvard pages were used instead.
• AST SpaceMobile produced no usable fresh output in today’s X rotation.
• Fresh IEEE and ACM discovery was lower-signal than the strongest new arXiv entries, so Research Radar prioritized the most relevant last-7-day arXiv papers.

Takeaway

The clearest pattern this morning is infrastructure hardening: AI is moving deeper into government codebases, enterprise workflows, telecom operations, and custom silicon instead of staying at the demo layer.

Daily Digest — Monday, July 6, 2026

Apptronik expands Robot Park and tightens its Gemini Robotics loop

Apptronik announced an expanded Robot Park in Austin and positioned Apollo 2 as the data-collection workhorse for a broader continuous-learning system. The key point is that fleets of humanoid robots are now being used to generate real-world training data at scale, not just to stage demos.

As part of Apptronik’s research partnership with Google DeepMind, data from Apollo 2 deployments helps train Gemini Robotics. That makes this one of the clearest recent examples of a robotics company building an industrialized feedback loop between physical deployments and foundation-model improvement.

Source: https://apptronik.com/news-collection/welcome-to-robot-park-where-apptroniks-apollo-goes-to-work

Amazon Leo passes 375 satellites in orbit

Amazon’s latest Leo mission update says the network now has more than 375 satellites in orbit after successive Atlas V and Ariane 6 launches. That doesn’t put it near Starlink scale, but it does show that Amazon’s LEO effort is now accumulating real orbital mass instead of remaining an early-stage launch story.

For wireless and satellite-watchers, the significance is competitive pressure: a second deep-pocketed broadband constellation is building cadence, launch diversity, and deployment muscle.

Source: https://www.aboutamazon.com/news/innovation-at-amazon/project-kuiper-satellite-rocket-launch-progress-updates

Emirates hits one million Starlink Wi-Fi connections

Emirates says customers have already made more than one million Starlink Wi-Fi connections since rollout began seven months ago. The airline also says it now operates more than 60 Starlink-equipped flights per day and has already seen more than a petabyte of customer usage.

That matters because it suggests inflight satellite internet is crossing from novelty into expected infrastructure, especially when a major global carrier can show rapid adoption at fleet scale.

Source: https://www.emirates.com/media-centre/one-million-connections-and-counting-emirates-customers-embrace-starlink-wi-fi/

Anthropic launches Claude Science for reproducible research workflows

Anthropic released Claude Science in beta for Pro, Max, Team, and Enterprise users. The product is designed to generate scientific artifacts such as figures and manuscripts alongside the code and environment that produced them, while also handling compute across laptops, clusters, and on-demand GPUs.

The interesting part for researchers is not just convenience but auditability. Anthropic is clearly aiming at a workflow where AI-generated scientific work stays inspectable and reproducible instead of becoming a black-box assistant.

Source: https://www.anthropic.com/news/claude-science-ai-workbench

Ericsson pushes AI into the radio access network itself

Ericsson introduced AI in RAN, a software subscription that brings telco-grade AI models directly into basebands and radios. The company says the system is already proven in more than 15 deployments and trials, with gains in throughput, spectral efficiency, user density handling, and positioning precision.

For Dad’s research lane, this is one of the more concrete signs that AI-native RAN is maturing from concept into deployable software that works within existing 5G Advanced hardware and Cloud RAN environments.

Source: https://www.ericsson.com/en/news/2026/6/the-ran-gets-smarter-ericsson-puts-ai-where-it-matters

GitHub sets a final shutdown date for GitHub Models

GitHub says GitHub Models will be fully retired on July 30, 2026. Before that, scheduled brownouts on July 16 and July 23 will temporarily break requests to help users prepare for shutdown.

This is worth watching because it reflects a broader platform trend: instead of supporting multiple overlapping AI surfaces, developer platforms are consolidating around narrower default experiences such as Copilot and partner ecosystems.

Source: https://github.blog/changelog/2026-07-01-github-models-is-being-fully-retired-on-july-30-2026/

NVIDIA claims strong ICML 2026 research footprint for Nemotron and GPUs

In a July 6 post, NVIDIA said 145 accepted ICML 2026 papers cite Nemotron models and datasets, and around 2,000 accepted papers cite NVIDIA GPUs. Even if the framing is self-promotional, the post is a notable signal that NVIDIA is pushing hard to make its open-model stack visible in mainstream academic research.

That matters because AI infrastructure leadership is increasingly being contested not just on chips, but on the surrounding software, models, datasets, and research mindshare.

Source: https://x.com/nvidia/status/2074161704456356216

Research Radar

Ultra-Low-Cost Hybrid Beamforming: A New Static-Connection Architecture with Sparse Phase-Shifter Sharing

Authors: Honghao Wang, Qingqing Wu, Yifei Wu, Yuxuan Chen, Wen Chen, Derrick Wing Kwan Ng
Venue: arXiv
This paper proposes a hybrid-beamforming architecture that reduces phase-shifter count while preserving most of the beamforming capability of conventional sub-connected designs. It looks practically relevant because it focuses on hardware-efficiency tradeoffs rather than only idealized performance.

Source: https://arxiv.org/abs/2607.02393v1

Three-Dimensional Spatial Correlation Modeling for Cylindrical mMIMO Arrays in HAPS

Authors: Shasha Liu, Abla Kammoun, Mohamed-Slim Alouini
Venue: arXiv
This work derives a closed-form spatial-correlation function for cylindrical massive-MIMO arrays on high-altitude platform stations. That makes it useful for future non-terrestrial network modeling where cylindrical geometries are more realistic than the planar assumptions used in many terrestrial studies.

Source: https://arxiv.org/abs/2607.02111v1

Robust Transmission Design for RIS-Assisted RSMA-SWIPT Systems With Movable Antennas Under Hardware Distortions

Authors: Muhammad Asif, Asim Ihsan, Irfan Muhammad, Mohd Hamza Naim Shaikh, Syed Tariq Shah, Zhu Shoujin, Symeon Chatzinotas
Venue: arXiv
This paper tackles a hard joint optimization problem involving movable antennas, RIS configuration, power splitting, and hardware impairments. It is dense, but it sits squarely in the broader trend of increasingly adaptive and software-defined wireless architectures.

Source: https://arxiv.org/abs/2607.02384v1

MIT/Harvard Events This Week

• July 7 — LL Technology Office Seminar: From Concept to Mission Impact @ MIT Virtual
  Source: https://calendar.mit.edu/event/ll-technology-office-seminar-engine-for-change
• July 7 — Founder Talk: Allison Byers of Scroobious @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/founder-talk-allison-byers-of-scroobious-author-of-fundraising-for-the-rest-of-us
• July 8 — Venture Incubation Program Advancement Ceremony @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/venture-incubation-program-advancement-ceremony
• July 9 — Virtual Founder Circle @ Online
  Source: https://innovationlabs.harvard.edu/events/virtual-founder-circle

Source Issues

• TNT’s event calendar was still stale and largely surfaced February–April listings, so current MIT and Harvard event pages were used instead.
• AST SpaceMobile and OneWeb did not yield usable fresh posts during this run.
• arXiv API requests began rate-limiting mid-run, so the paper shortlist was assembled from successfully fetched fresh results.

Takeaway

The clearest through-line today is that advanced tech systems are becoming operational systems: humanoid robots are collecting production data, satellite networks are proving real adoption at scale, telecom AI is moving into the RAN, and research assistants are being built around reproducibility instead of novelty alone.

Morning roundup for Friday, July 3, 2026.

Top Stories

Anthropic publishes a cyber-jailbreak severity framework for Fable 5

Anthropic released a more explicit explanation of the cyber safety classifiers that ship with Claude Fable 5, along with a draft framework for scoring AI jailbreak severity. The interesting part is not just the safeguards themselves, but the attempt to standardize how labs, governments, and security researchers talk about prompt-based failures and misuse risk.

Source: https://www.anthropic.com/news/fable-safeguards-jailbreak-framework

Google’s latest AI roundup points toward ambient agents

Google’s July 1 recap pulls together several product directions that matter more in combination than in isolation: Gemma 4 12B running locally, computer use in Gemini 3.5 Flash, and Gemini 3.5 Live Translate across more than 70 languages. Taken together, the company is clearly trying to make AI feel like an always-available layer across devices, voice, and browser workflows.

Source: https://blog.google/innovation-and-ai/technology/ai/google-ai-updates-june-2026/

AWS expands Bedrock AgentCore into four more regions

AWS said Bedrock AgentCore is now available in Bangkok, Malaysia, Milan, and Spain. Regional spread matters here because agent systems become much more viable in production once runtime, persistence, policy, and tool access can run closer to end users with lower latency.

Source: https://aws.amazon.com/about-aws/whats-new/2026/06/amazon-bedrock-agentcore-four-additional-regions/

AWS raises AgentCore’s default runtime quotas

AWS also increased Bedrock AgentCore’s default capacity, including up to 5,000 active concurrent sessions in US East and US West and 200 agent interactions per second in every supported region. That is a notable infrastructure signal: cloud providers are now tuning agent platforms for sustained multi-session production load rather than occasional experiments.

Source: https://aws.amazon.com/about-aws/whats-new/2026/07/amazon-bedrock-agentcore-increases-default-runtime-quota-limits/

Qualcomm and Hugging Face deepen their hybrid AI push

Qualcomm expanded its relationship with Hugging Face around open, developer-driven AI spanning edge devices and data-center systems. The release emphasizes agent orchestration across device and cloud, which reinforces the broader industry move toward hybrid inference based on cost, latency, privacy, and performance.

Source: https://www.qualcomm.com/news/releases/2026/06/qualcomm-and-hugging-face-expand-relationship-to-advance-open--d

The FCC advances Upper C-band policy for more mid-band wireless capacity

A July 1 FCC order adds a primary non-federal mobile allocation in 4.0–4.16 GHz and establishes a 20 MHz guard band to support coexistence with remaining satellite services. For anyone tracking 5G Advanced and 6G foundations, this is another reminder that spectrum policy remains one of the most important levers shaping actual network capability.

Source: https://docs.fcc.gov/public/attachments/DOC-422738A1.pdf

NVIDIA keeps pushing physical AI through reusable robotics software

NVIDIA’s latest robotics-focused piece highlights Isaac ROS as a software layer for perception, mapping, collision detection, and motion planning across mobile robots, manipulators, and humanoids. The key idea is that physical AI may scale fastest when the software stack becomes modular, open, and deployable across many embodiments.

Source: https://blogs.nvidia.com/blog/nvidia-life-jaiveer-singh/

Nokia joins Finland’s nationwide counter-drone initiative

Nokia Defense said it is participating in a Finnish Border Guard-led consortium to develop next-generation counter-UAS capabilities for patrol vehicles and boats. The release shows how defense connectivity is increasingly being packaged as an integrated stack of sensing, secure communications, interoperability, and real-time response.

Source: https://www.nokia.com/newsroom/nokia-to-provide-intelligent-connectivity-for-finnish-border-guard-counter-drone-initiative-nationwide/

Research Radar

Opportunistic Positioning with LEO Satellites based on SSB from NR NTN

Authors: Rainer Bachl, Muhammad Nabeel, Tingting Lei
Venue: arXiv
This paper studies how synchronization signal blocks from future NR NTN LEO systems could support opportunistic positioning, which is highly relevant to satellite-assisted mobility and timing.

Source: http://arxiv.org/abs/2607.00967v1

Scalable Security and Migration-Aware SFC Provisioning in LEO Satellite Networks

Authors: Mohammed Mahyoub, Wael Jaafar, Sami Muhaidat, Halim Yanikomeroglu
Venue: arXiv
The paper proposes service-function-chain provisioning for multi-tenant LEO satellite networks while explicitly accounting for security virtual network functions and satellite mobility.

Source: http://arxiv.org/abs/2607.00471v1

Enabling Real-Time AI in O-RAN: Deploying and Measuring AI Inside a Near-RT RIC xApp

Authors: Lawrence Obiuwevwi, Krzysztof J. Rechowicz, Sampath Jayarathna, Safdar Hussain Bouk
Venue: arXiv
A practical O-RAN paper that focuses on deploying and measuring AI inference inside a near-real-time RIC xApp instead of only evaluating offline or in simulation.

Source: http://arxiv.org/abs/2607.01583v1

MIT / Harvard Events This Week

• July 6 — Writing Together Online: Keep Your Writing Momentum This Summer @ MIT Virtual
  Source: https://calendar.mit.edu/event/copy-of-writing-together-online-with-the-wcc-get-it-done-in-june
• July 6 — Music Theory Pedagogies Reimagined: Conference and Hackathon @ MIT Building W18, Tull Concert Hall
  Source: https://calendar.mit.edu/event/music-theory-pedagogies-reimagined
• July 7 — LL Technology Office Seminar: From Concept to Mission Impact @ MIT Virtual
  Source: https://calendar.mit.edu/event/ll-technology-office-seminar-engine-for-change
• July 7 — Founder Talk: Allison Byers of Scroobious @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/founder-talk-allison-byers-of-scroobious-author-of-fundraising-for-the-rest-of-us
• July 8 — Venture Incubation Program Advancement Ceremony @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/venture-incubation-program-advancement-ceremony
• July 9 — Virtual Founder Circle @ Online
  Source: https://innovationlabs.harvard.edu/events/virtual-founder-circle

Source Issues

• TNT’s calendar page was stale and mostly surfaced February–April events, so current MIT and Harvard pages were used instead.
• Direct arXiv page fetches returned 400 errors during this run, so paper discovery fell back to the arXiv API.
• IEEE and ACM searches were lower-signal than fresh arXiv results for today’s research bands, so the digest prioritized stronger recent arXiv papers.

Takeaway

The strongest signal this morning is that AI agents are being operationalized from every angle at once: safety frameworks, cloud runtime quotas, regional availability, hybrid edge-cloud orchestration, and the telecom/spectrum substrate underneath them.

Morning Digest for Thursday, July 2, 2026

This morning’s pattern is operationalization: AI labs, telecom vendors, chip companies, and satellite operators are all pushing research and infrastructure into live, continuous deployment.

Top Stories

OpenAI launched GeneBench-Pro for AI biological research workflows

OpenAI introduced GeneBench-Pro as a research-level benchmark designed to test whether agents can work through messy biological data, choose appropriate analytical paths, and make the kinds of judgment calls real computational biology depends on. That makes it more interesting than a standard benchmark release because it evaluates workflow competence, not just neat question-answer performance.

Source: https://x.com/OpenAI/status/2072004836674167294

NVIDIA is pitching multi-tenant AI factories as the next compute business model

NVIDIA said it is partnering with AI clouds on revenue-sharing and credit-support structures to deploy large-scale, multi-tenant AI factories for startups, model builders, enterprises, and regional AI operators. The strategic shift here is that NVIDIA is increasingly selling an operating and financing model for always-on token production, not merely hardware.

Source: https://x.com/nvidia/status/2072545807505527251

Qualcomm and Meta are collaborating on next-generation data-center CPUs for AI

Qualcomm highlighted a collaboration with Meta around next-generation data-center CPUs built for AI infrastructure. This reinforces that Qualcomm’s AI ambition now extends materially beyond mobile and edge, and that hyperscale AI compute is attracting a much wider semiconductor field.

Source: https://x.com/Qualcomm/status/2072390083655704824

SpaceX launched and deployed another 24 Starlink satellites from California

SpaceX confirmed a new Falcon 9 mission that launched and deployed 24 Starlink satellites from California on July 2. While Starlink launches are becoming routine, that routine itself is the story: LEO broadband deployment is now running at infrastructure scale.

Source: https://x.com/SpaceX/status/2072561934134624613

Starlink is supporting air-ambulance connectivity across remote Philippine islands

Starlink said air ambulance crews in the Philippines are using the network to coordinate with medical teams, monitor weather, and make in-flight operational decisions across remote islands. This is a strong example of satellite internet moving into mission-critical healthcare operations.

Source: https://x.com/Starlink/status/2072368009776414860

Ericsson says RAN automation is moving toward AI systems that decide without human intervention

Ericsson’s latest RAN automation guide argues that the next phase of network operations will depend on AI-powered systems capable of making the right decisions without manual intervention. For wireless researchers, this is another concrete industry signal that AI-native RAN control is becoming a real deployment target.

Source: https://x.com/ericsson/status/2072606210226897252

Jim Fan introduced ASPIRE, a compounding skill library for robots

Jim Fan described ASPIRE as a framework where coding agents evolve reusable robot skills over time, building up a software-like library of know-how rather than repeatedly retraining one end-to-end policy from scratch. The big idea is that robot learning may become more cumulative, modular, and transferable across embodiments.

Source: https://x.com/DrJimFan/status/2072004190856212902

Research Radar

LEO Satellite Network Orchestration with Heterogeneous Graph Neural Networks

Authors: Aruna Jayarajan, N. Cameron Matson, Karthikeyan Sundaresan
Venue: arXiv
This paper proposes NEO-GNN, a heterogeneous graph neural network for real-time association decisions across satellites, gateways, and ground cells in bent-pipe LEO systems. It is directly relevant to balancing coverage and traffic satisfaction in highly dynamic satellite-ground networks.

🔗 https://arxiv.org/abs/2606.31950v1

An LLM-Based Framework for Intent-Driven Network Topology Design

Authors: Kholoud El-Habbouli, Fen Zhou, Stephane Huet
Venue: arXiv (submitted to IEEE CNSM 2026)
This work evaluates whether LLMs can convert natural-language network requirements into constraint-compliant and resilient topologies. It is especially interesting as a benchmark for AI-assisted network design rather than generic chatbot performance.

🔗 https://arxiv.org/abs/2607.00292v1

SNR-Adaptive Optimal Threshold Design for Energy Detection in Dynamic Spectrum Access

Authors: Sushila Dhaka, Jane-Hwa Huang, Chin-Min Yu, Li-Chun Wang
Venue: IEEE VTC 2026 Spring Workshop / arXiv
The paper derives a closed-form adaptive thresholding approach for energy detection under heterogeneous SNR conditions, reducing sensing error relative to fixed-threshold methods. That makes it a neat fit for spectrum-sharing and secure cooperative sensing discussions.

🔗 https://arxiv.org/abs/2607.00754v1

MIT/Harvard Events This Week

• July 2 — Founder/Joiner Connections: Pizza + Pitch @ Harvard Innovation Labs
  🔗 https://innovationlabs.harvard.edu/events/founder-joiner-connections-pizza-pitch
• July 7 — LL Technology Office Seminar: From Concept to Mission Impact @ MIT Virtual
  🔗 https://calendar.mit.edu/event/ll-technology-office-seminar-engine-for-change
• July 7 — Founder Talk: Allison Byers of Scroobious @ Harvard Innovation Labs
  🔗 https://innovationlabs.harvard.edu/events/founder-talk-allison-byers-of-scroobious-author-of-fundraising-for-the-rest-of-us

Source Notes

• TNT’s calendar page remained stale and mostly listed February–April events, so current MIT and Harvard pages were used instead.
• AST SpaceMobile returned no recent posts during this run, so it was skipped.
• ACM search was blocked by anti-bot protection and IEEE’s search page was low-signal in fetch mode, so the research section prioritized the strongest fresh arXiv and IEEE-linked papers.
• Several rotated X accounts overlapped with the previous three digests and were excluded to avoid repetition.

Takeaway

The clearest pattern this morning is that AI and networking progress is no longer just about announcing models or architectures—it is about building systems that can operate continuously in production, from satellite links and network automation to scientific agents and robot skill transfer.

Daily Digest — July 1, 2026

Anthropic brings Claude Fable 5 back globally with tighter abuse controls

Anthropic said Claude Fable 5 will return globally after discussions with the U.S. government, with new classifiers designed to block more cybersecurity misuse. In the near term, some routine coding and debugging tasks will fall back to Opus 4.8 while Anthropic refines the filters.

This matters because it shows frontier-model deployment increasingly happening under live regulatory and national-security pressure rather than purely lab-internal release cycles. Model access, safeguards, and government coordination are becoming part of the product surface.

Source: https://x.com/AnthropicAI/status/2072163884430229756

OpenAI unveils its first in-house AI chip, Jalapeño

OpenAI announced Jalapeño, its first internally designed AI chip, built with Broadcom and aimed at the LLM workloads behind ChatGPT, Codex, the API, and future agentic systems.

The strategic signal is obvious: OpenAI is moving farther down the stack into infrastructure control. If that direction holds, the company is no longer just a model and application lab but a vertically integrated AI platform player.

Source: https://x.com/OpenAI/status/2069770172802773292

Google DeepMind ships Nano Banana 2 Lite and Gemini Omni Flash

Google DeepMind announced two multimodal releases: Nano Banana 2 Lite for cheaper, faster image generation, and Gemini Omni Flash for video generation and editing via the Gemini API and Google AI Studio.

The bigger implication is workflow composition. Google is pushing a stack where developers can generate an image, preserve session state, and immediately animate or edit it inside a unified tool chain.

Source: https://x.com/GoogleDeepMind/status/2071988044878516466

NVIDIA says software alone cut DeepSeek V4 token costs sharply on Blackwell

NVIDIA said software optimizations on Blackwell improved DeepSeek V4 inference performance by up to 5× in one month, reducing token costs to roughly one-fifth of prior levels.

That is a useful reminder that inference economics are no longer just a hardware story. Runtime, kernels, networking, and systems integration are becoming persistent levers for lowering cost after infrastructure is already deployed.

Source: https://x.com/nvidia/status/2071979909199577560

Qualcomm points to a fresh 6G standardization milestone at 3GPP

Qualcomm Research said the latest 3GPP RAN plenary marked a major milestone for 6G standardization.

For wireless researchers, the important part is not the marketing language but the timing: 6G is moving closer to concrete protocol and study-item framing, which means more of the discussion will shift from vision slides to technical implementation constraints.

Source: https://x.com/Qualcomm/status/2072080143724711969

Starlink Mobile expands across the Philippines with Globe

Starlink said Globe customers across the Philippines’ 7,000+ islands can now use apps and messaging through Starlink Mobile.

This is a meaningful direct-to-cell datapoint because the Philippines is exactly the kind of fragmented geography where satellite-mobile integration can solve structural terrestrial coverage gaps rather than merely add novelty.

Source: https://x.com/Starlink/status/2071984086626025689

Nokia highlights AI-driven programmable 5G for drone operations

Nokia said AI-driven programmable networks, ultra-low latency, massive IoT, and intelligent orchestration are enabling connected drone operations with Telefónica.

The industry pattern here is that telecom vendors are reframing 5G as an execution layer for autonomous systems. The sale is no longer just radio access; it is orchestration, latency guarantees, and machine coordination.

Source: https://x.com/nokia/status/2071873533953663324

Research Radar

Semantic-Aware Multiple Access via Spatial Redundancy Exploitation for Uplink-Dominant 6G Use Cases

Authors: Hamidreza Mazandarani, Masoud Shokrnezhad, Tarik Taleb, Onur Günlü
Venue: arXiv

This paper proposes a 6G uplink design that exploits overlapping fields of view among nearby devices so they do not all send the same semantic content. For dense sensing or vehicular environments, that is a practical path toward reducing uplink waste while preserving reconstruction quality.

Source: https://arxiv.org/abs/2606.31715v1

Towards a Joint Task-Oriented and Generative Semantic Communication Framework for 6G Networks

Authors: Soheyb Ribouh, Phil Polo Ditsia Di Ngoma
Venue: arXiv

This work combines semantic compression with both downstream inference and scene reconstruction, showing how 6G communication systems might support control tasks and human-usable output simultaneously. It is a nice example of semantic communications being treated as a system design question rather than a single compression trick.

Source: https://arxiv.org/abs/2606.31426v1

Toward AI-Native 6G Air Interface: A 3GPP Perspective on Protocol Framework

Authors: Xingqin Lin
Venue: arXiv

This paper argues that 6G should standardize the protocol framework around AI-enabled behavior rather than over-prescribe model architectures, training methods, or weights. That distinction matters if vendors want interoperability without freezing innovation too early.

Source: https://arxiv.org/abs/2606.27466v1

MIT/Harvard Events This Week

• July 1 — SPUR / RSI 2026 Lecture Series @ MIT Building 2, Room 190
  Source: https://calendar.mit.edu/event/spur-rsi-2026-lecture-series
• July 2 — Founder/Joiner Connections: Pizza + Pitch @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/founder-joiner-connections-pizza-pitch
• July 7 — LL Technology Office Seminar: From Concept to Mission Impact @ MIT Virtual
  Source: https://calendar.mit.edu/event/ll-technology-office-seminar-engine-for-change
• July 7 — Founder Talk: Allison Byers of Scroobious @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/founder-talk-allison-byers-of-scroobious-author-of-fundraising-for-the-rest-of-us

Source Issues

• TNT’s calendar page was stale and mostly showed February–April entries, so current MIT and Harvard event pages were used instead.
• IEEE and ACM searches were checked, but the strongest fresh research matches for Dad’s topic bands were on arXiv this morning.
• Several rotated X accounts were quiet or overlapped with stories already covered in the prior three digests, so they were skipped to preserve freshness.

Takeaway

The strongest pattern this morning is vertical integration: AI labs, chip vendors, telecom firms, and satellite operators are all trying to own more of the path from capability to deployment.

☀️ Morning Digest — Tuesday, June 30

1) Palantir brings NVIDIA Nemotron into air-gapped government and critical-infrastructure deployments

NVIDIA says Palantir’s new intelligent engine uses Nemotron open models inside isolated environments, giving U.S. agencies and critical operators a path to sovereign AI without opening sensitive networks. The broader strategic signal is that open frontier models are now being positioned for closed, high-trust deployments rather than only public-cloud experimentation.

Source: https://blogs.nvidia.com/blog/palantir-secure-ai-us-agencies-nemotron-open-models/

2) Anthropic’s Claude is now generally available in Microsoft Foundry on NVIDIA GB300 systems

NVIDIA says Claude running on Azure-hosted GB300 NVL72 infrastructure is now generally available for enterprises building autonomous and domain-specific agents. This looks like another step toward vertically integrated enterprise agent stacks where compute, networking, security, and model access are sold together.

Source: https://blogs.nvidia.com/blog/anthropic-nvidia-gb300-blackwell-ultra-microsoft-azure/

3) Google is expanding Gemini’s personalized image creation to eligible U.S. users

Google says Gemini can now use opt-in context from tools like Google Photos, Gmail, YouTube, and Search to create more personalized images for free in the U.S. The product implication is that consumer AI is shifting from generic generation toward assistant-style outputs grounded in a user’s actual data and preferences.

Source: https://blog.google/innovation-and-ai/products/gemini-app/personal-intelligence-nano-banana-us-expansion/

4) Nokia says AI-RAN is shifting from industry concept to commercialization

In a new Nokia post, the company argues telecom is moving from AI-RAN experimentation toward commercial deployment as operators prepare for deterministic, AI-heavy workloads. For wireless researchers, the key angle is that AI is increasingly being treated as a first-class network workload rather than just a tool for optimizing legacy traffic.

Source: https://www.nokia.com/blog/accelerating-ai-ran-from-concept-to-commercialization/

5) SpaceX launched another 24 Starlink satellites from California

SpaceX confirmed a Falcon 9 launch and deployment of 24 Starlink satellites on June 28, continuing its high-cadence expansion of LEO broadband capacity. What matters most is the consistency: satellite deployment is now operating with the rhythm of infrastructure, not special-event launches.

Source: https://x.com/SpaceX/status/2071303632880808132

6) Ericsson is pushing agentic AI deeper into autonomous network operations

Ericsson says service orchestration for autonomous operations now depends on agentic AI, modular ODA-aligned components, and real-time inventory. That aligns with the wider telecom vendor push to turn OSS/BSS modernization into the foundation for self-operating networks.

Source: https://x.com/ericsson/status/2071636122183082014

📡 Research Radar

Direct-to-Cell: A First Look into Starlink’s Direct Satellite-to-Device Radio Access Network through Crowdsourced Measurements

Authors: Jorge Garcia-Cabeza, Javier Albert-Smet, Zoraida Frias, Luis Mendo, Santiago Andrés Azcoitia, Eduardo Yraola
Venue: arXiv

This measurement study is especially relevant because it moves beyond vendor claims and uses field data to characterize early direct-to-cell behavior. Its beam-level capacity estimates and signal-quality observations make it useful context for thinking about realistic DS2D performance limits.

Source: https://arxiv.org/abs/2506.00283

Efficient Self-Learning and Model Versioning for AI-native O-RAN Edge

Authors: Mounir Bensalem, Fin Gentzen, Tuck-Wai Choong, Yu-Chiao Jhuang, Admela Jukan, Jenq-Shiou Leu
Venue: arXiv

This paper focuses on the operational challenge of managing many ML models across O-RAN control loops and heterogeneous edge/cloud layers. The closed-loop update and versioning angle makes it particularly relevant for real deployments of xApps, rApps, and AI-native control stacks.

Source: https://arxiv.org/abs/2601.17534

Federated Learning-driven Beam Management in LEO 6G Non-Terrestrial Networks

Authors: Maria Lamprini Bartsioka, Ioannis A. Bartsiokas, Athanasios D. Panagopoulos, Dimitra I. Kaklamani, Iakovos S. Venieris
Venue: arXiv

The paper studies federated beam selection for LEO NTNs and shows how distributed learning can improve prediction stability under changing orbital and channel conditions. That makes it a good fit for the growing overlap between 6G intelligence and non-terrestrial network control.

Source: https://arxiv.org/abs/2603.10983

🎓 MIT/Harvard Events This Week

• June 30 — LL Technology Office Seminar: From 3 Configurations to 300 @ MIT Virtual
  Source: https://calendar.mit.edu/event/ll-technology-office-seminar-nTop-Bradley-Rothenberg
• June 30 — Summer Seminar Series on Super Pollutants @ MIT Building 55-110
  Source: https://calendar.mit.edu/event/summer-seminar-series-on-super-pollutants
• July 1 — SPUR / RSI 2026 Lecture Series @ MIT Building 2, Room 190
  Source: https://calendar.mit.edu/event/spur-rsi-2026-lecture-series
• July 2 — Founder/Joiner Connections: Pizza + Pitch @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/founder-joiner-connections-pizza-pitch

⚠️ Source Issues

• TNT’s calendar page was still stale and mostly surfaced February–April listings, so current MIT and Harvard event pages were used instead.
• AST SpaceMobile’s feed returned no recent posts during this run, so it was skipped rather than forcing a weak item.
• Recent arXiv results in Dad’s exact topic bands were noisy and overlapped with the past three digests, so the radar kept the strongest non-duplicative papers even when some were slightly older than the ideal 7-day window.

💡 Takeaway

This morning’s clearest pattern is operationalization: AI and telecom players are packaging agents, infrastructure, and network control into systems designed to run in production, not just in demos.

Morning Digest

HP expands OpenAI Frontier from pilot projects into broader enterprise deployment

OpenAI says HP is scaling its Frontier strategic partnership after early pilots showed value in engineering, software remediation, and productivity workflows. The interesting part is not the existence of another enterprise AI deployment, but the fact that HP is treating Frontier as a connective operating layer for access, context, governance, evaluation, and agent deployment across customer support, partner operations, telemetry, and internal software work.

Source: https://openai.com/index/hp-frontier-partnership/

Qualcomm debuts Dragonfly as a data-center portfolio for the agentic AI era

At Qualcomm Investor Day, the company introduced the Dragonfly brand and positioned it around the C1000 CPU, AI300 accelerators, data center DSPs, and a near-memory computing architecture. The larger strategic read is that Qualcomm is no longer presenting itself as mainly a device company with adjacent AI ambitions; it is explicitly trying to become a cloud-and-device infrastructure player for agentic workloads.

Source: https://x.com/Qualcomm/status/2069895327004475525

Qualcomm moves to acquire Modular to strengthen its software layer

Qualcomm also announced an agreement to acquire Modular, aiming to expand its AI software capabilities across edge and data center environments. Taken together with Dragonfly, this looks like a full-stack move: the company wants not only silicon relevance, but a stronger position in the software and developer tooling layer that determines where AI workloads actually land.

Source: https://www.qualcomm.com/news/releases/2026/06/qualcomm-to-acquire-modular

SpaceX successfully launched and deployed SiriusXM’s SXM-11 satellite

SpaceX confirmed Falcon 9 lifted off from Florida and later deployed the SXM-11 mission successfully. By itself, this is a straightforward commercial launch milestone, but the repeated pattern matters more: SpaceX continues to turn what used to be exceptional orbital delivery into a normalized, high-frequency logistics service.

Source: https://x.com/SpaceX/status/2071428922705657956

Starlink Mobile turned on free SMS connectivity for MovistarVe users in earthquake-hit Venezuela

Starlink said compatible LTE smartphones in La Guaira can now automatically connect for messaging, even if terrestrial infrastructure is unavailable, while it works to extend support to additional carriers. This is a notable direct-to-cell development because it shows satellite-mobile integration moving from announcement-stage potential into an emergency-response communications tool that can function on ordinary phones.

Source: https://x.com/Starlink/status/2071236407738384869

NVIDIA spotlighted a local-first enterprise AI workflow at Zaha Hadid Architects

NVIDIA highlighted how Zaha Hadid Architects is using local compute, fine-tuned AI models, and NVIDIA technologies to speed up design workflows while keeping proprietary data secure. That is a useful signal for enterprise AI adoption more broadly: deployment decisions are increasingly being shaped by governance, data residency, and trust requirements rather than by raw benchmark capability alone.

Source: https://x.com/nvidia/status/2070597978826449174

Research Radar

Parameter Efficient Hybrid Transformer (PEHT) for Network Traffic Prediction via Dynamic Urban Congestion Integration

Authors: Abdolazim Rezaei, Mehdi Sookhak, Mahboobeh Haghparast
Venue: arXiv
PEHT combines cellular traffic features with urban mobility and congestion signals, using a parameter-efficient Transformer design to improve prediction quality without bloating training cost. For wireless systems work, it is interesting because it treats city dynamics as first-class network inputs rather than background noise.

🔗 https://arxiv.org/abs/2606.28274v1

Adaptive Machine Learning Framework for UAV Trajectory Optimization in O-RAN

Authors: Chenrui Sun, Swarna Bindu Chetty, Gianluca Fontanesi, Mahnaz Arvaneh, Walid Saad, Hamed Ahmadi
Venue: IEEE Transactions on Vehicular Technology / arXiv
This paper proposes continual transfer learning inside O-RAN to speed up UAV trajectory adaptation in new environments. That matters because 6G-style airborne coverage systems will only be practical if they can adapt quickly without expensive retraining every time the scene changes.

🔗 https://arxiv.org/abs/2606.24483v1

Joint Visibility Analysis of RIS in Non-Terrestrial Networks through Stochastic Geometry

Authors: Ashutosh Balakrishnan, Junse Lee, Francois Baccelli
Venue: arXiv
The paper analyzes when rooftop RIS deployments are jointly visible to both user equipment and non-terrestrial base stations in urban settings. It is a good reminder that NTN performance is not just about satellites and spectrum; geometry and blockage statistics remain central design constraints.

🔗 https://arxiv.org/abs/2606.21946v1

MIT/Harvard Events This Week

• June 29 — Group Office Hours: Customer Discovery @ Online
  Source: https://innovationlabs.harvard.edu/events/group-office-hours-customer-discovery

• June 30 — LL Technology Office Seminar: From 3 Configurations to 300 @ MIT Virtual
  Source: https://calendar.mit.edu/event/ll-technology-office-seminar-nTop-Bradley-Rothenberg

• June 30 — Summer Seminar Series on Super Pollutants @ MIT Building 55-110
  Source: https://calendar.mit.edu/event/summer-seminar-series-on-super-pollutants

• July 2 — Founder/Joiner Connections: Pizza + Pitch @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/founder-joiner-connections-pizza-pitch

Source Notes

• TNT’s calendar page was stale and mostly surfaced February–April listings, so current MIT and Harvard event pages were used instead.
• Several rotated X accounts were skipped because they were either quiet or overlapped with stories already covered in the last three digests.
• Qualcomm’s release index was only partially readable via fetch, so this digest used a mix of the official acquisition page and the company’s own X posts for Investor Day details.
• arXiv search remained noisy, so the paper list was manually filtered toward wireless, O-RAN, and non-terrestrial-network relevance.

Takeaway

Today’s clearest pattern is convergence: frontier AI, telecom hardware, and satellite systems are all being reorganized into integrated deployment stacks rather than isolated products.

Morning Digest — Sunday, June 28

OpenAI says agents are becoming the default way work gets done inside the company

OpenAI published a new report on June 25 arguing that long-running agents are no longer a side workflow inside the company. Codex now appears to dominate internal AI usage, not just for engineers but also for legal, recruiting, finance, and operations teams. The meaningful shift is structural: agents are being treated less like assistants and more like delegated labor that can run for hours across parallel tasks.

Source: https://openai.com/index/how-agents-are-transforming-work/

Anthropic’s latest Economic Index maps how AI usage now follows the rhythms of daily work

Anthropic’s June 2026 Economic Index report adds hourly sampling and survey data to show how Claude use changes across the day, surfaces, and occupations. That makes the conversation around AI’s labor impact much more concrete, because it starts grounding claims in observed behavioral patterns instead of broad top-down assumptions.

Source: https://www.anthropic.com/research/economic-index-june-2026-report

Google DeepMind and A24 launched a research partnership around creative AI tools

Google announced a long-term research and development partnership between Google DeepMind and A24 spanning multiple projects. The interesting part is not just the announcement itself, but the model of collaboration: frontier labs are increasingly trying to co-design creative tools with actual studios and creators rather than shipping generic capability and hoping the industry adapts.

Source: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/deepmind-a24-research-partnership/

Google DeepMind published its AI Control Roadmap for securing increasingly capable agents

DeepMind’s new roadmap frames agent security as a layered problem across models, multi-agent systems, and the broader cyber-defense ecosystem. That suggests leading labs are preparing for a near future where advanced agents are useful enough to deploy widely but still imperfect enough that operational containment and monitoring become essential infrastructure.

Source: https://deepmind.google/blog/securing-the-future-of-ai-agents/

SpaceX completed a full-duration single-engine static fire test of Starship

On June 26, SpaceX posted that Starship had completed a full-duration single-engine static fire test. It is not a flashy milestone compared with a launch, but it is exactly the kind of ground-test progression that signals movement toward the next integrated campaign.

Source: https://x.com/SpaceX/status/2070482358369763674

Starlink, 4-H, and Tractor Supply will expand broadband access for underserved rural communities

Starlink announced a collaboration with 4-H and Tractor Supply to bring high-speed internet and online learning access to rural communities across the United States. That is notable because satellite connectivity is being positioned less as a premium alternative and more as public-interest infrastructure for education and STEM access.

Source: https://x.com/Starlink/status/2070931249431265606

Starlink Mobile is coming to Mongolia through GMobile

Starlink said its direct-to-cell service will soon be available in Mongolia through GMobile, aimed at herders, miners, and field workers beyond conventional network coverage. This is exactly the kind of incremental operator partnership that shows how satellite-mobile convergence is expanding in practice.

Source: https://x.com/Starlink/status/2070572849455333846

Research Radar

Wireless Personal Agent: Extending Wireless Intelligence from Networks to Terminals

Authors: Jiedan Tan, Fang Liu, Jingwen Tong, Shengli Zhang, Jun Zhang, Wing Shing Wong
Venue: arXiv
This paper argues for extending wireless intelligence from the network to the terminal itself via a personal-agent paradigm. That is interesting because it shifts optimization and adaptation closer to the user edge rather than keeping intelligence centralized in the infrastructure.

Source: https://arxiv.org/abs/2606.23255v1

Radio Resource Allocation for Beam Hopping Scheduling in LEO Satellite Communications: A Spatio-Temporal Perspective

Authors: Hao Yuan, Lanyining Li, Jianghua Long, Xing Zhang
Venue: arXiv
The paper studies beam-hopping scheduling with an explicit spatio-temporal view of resource allocation, which is directly relevant to how future LEO systems can scale coverage and capacity under dynamic demand.

Source: https://arxiv.org/abs/2606.22900v1

Cross-Layer Intrusion Detection in 5G O-RAN: Gains and Limits of Fusing Radio Telemetry with Network Flow Records

Authors: Hamed Fard, Ilya Komarov, Gerhard Wunder
Venue: arXiv
This work explores whether security systems gain meaningful detection power by fusing radio telemetry with network flow data in O-RAN environments. It is a useful paper for thinking about realistic defense architectures in AI-assisted, software-defined mobile networks.

Source: https://arxiv.org/abs/2606.22450v1

MIT/Harvard Events This Week

• June 29 — Group Office Hours: Customer Discovery @ Online
  Source: https://innovationlabs.harvard.edu/events/group-office-hours-customer-discovery
• June 30 — LL Technology Office Seminar: From 3 Configurations to 300 @ MIT Virtual
  Source: https://calendar.mit.edu/event/ll-technology-office-seminar-nTop-Bradley-Rothenberg
• June 30 — Summer Seminar Series on Super Pollutants @ MIT Building 55-110
  Source: https://calendar.mit.edu/event/summer-seminar-series-on-super-pollutants
• July 2 — Founder/Joiner Connections: Pizza + Pitch @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/founder-joiner-connections-pizza-pitch

Source Issues

• TNT’s calendar page was stale and mainly surfaced February–April listings, so I used direct MIT and Harvard event pages instead.
• Several rotated X accounts were quiet or overlapped with stories already covered in the last three digests, so I skipped them rather than force low-quality repeats.
• arXiv’s broad keyword search remained noisy, so I filtered recent results manually toward Dad’s wireless, LEO, and O-RAN interests.

Takeaway

The clearest pattern this morning is operationalization: AI agents are being embedded into real workflows, and satellite-plus-wireless systems are steadily becoming ordinary infrastructure instead of niche experiments.

Morning Digest — Saturday, June 27

OpenAI previewed GPT-5.6 Sol, Terra, and Luna

OpenAI began a limited preview of the GPT-5.6 family on June 26, led by Sol as the flagship model, Terra as the balanced lower-cost tier, and Luna as the fast high-volume tier. The big takeaway is not just stronger raw capability, but clearer segmentation of frontier, everyday, and scalable inference use cases, along with a more explicit safety-and-government coordination layer around launch.

Source: https://openai.com/index/previewing-gpt-5-6-sol/

Anthropic says Mythos 5 is being restored for some critical-infrastructure defenders

Anthropic said on June 27 that the U.S. government notified the company it could redeploy Mythos 5 to a set of U.S. organizations that operate and defend critical infrastructure. This is worth watching because it suggests top-end cyber-capable models are increasingly being governed like strategic infrastructure rather than released as ordinary SaaS products.

Source: https://x.com/AnthropicAI/status/2070665903440871779

Samsung is rolling out ChatGPT Enterprise and Codex across major parts of the company

OpenAI says Samsung Electronics is deploying ChatGPT Enterprise and Codex to all employees in Korea and all Device eXperience employees worldwide. The significance is scale: this looks less like a sandboxed AI experiment and more like a company-wide workflow platform spanning R&D, manufacturing, software, and corporate functions.

Source: https://openai.com/index/samsung-electronics-chatgpt-codex-deployment/

Qwen introduced AgentWorld as a native language world model for agents

Qwen says AgentWorld simulates seven agent environments—including terminal, web, OS, Android, MCP, search, and SWE settings—inside a single model. The architectural bet is important: training models to predict and model environments may become a serious alternative to simply scaling post-hoc tool-using agents.

Source: https://x.com/Alibaba_Qwen/status/2069720365442719867

Amazon Leo crossed 367 satellites in orbit and set its next launch for July 2

Amazon’s mission update says its June 17 Ariane 6 launch took the Leo constellation to 367 deployed spacecraft, and the next 29-satellite Atlas V mission is targeted for July 2, 2026. For the LEO broadband race, the interesting signal is launch cadence—Amazon is increasingly operating like a deployment machine rather than a project hitting isolated milestones.

Source: https://www.aboutamazon.com/news/innovation-at-amazon/project-kuiper-satellite-rocket-launch-progress-updates

NVIDIA is pushing agentic AI from demos into telecom operations

At DTW Ignite 2026, NVIDIA and partners are highlighting a stack built around synthetic telecom data, secure agent runtimes, telecom-domain models, and simulation for long-running operational agents. For anyone watching AI-for-networking, this is one of the clearest production-oriented attempts to turn agentic AI into telecom operations infrastructure.

Source: https://blogs.nvidia.com/blog/telecom-ai-agents-dtw-ignite-2026/

Research Radar

Maximum Achievable Burst Size in All-Optical Satellite Networks

Authors: Thomas Roethig, Sleman Mouammar, Ítalo Brasileiro, Soheil Hosseini, Admela Jukan
Venue: arXiv
This paper analyzes how burst-size ceilings vary across all-optical satellite constellations, which matters for understanding traffic engineering and transport-layer design in future high-capacity LEO backbones.

🔗 https://arxiv.org/abs/2606.27050v1

Toward Agentic SysAdmin: Rethinking System Administration with AI Agents

Authors: Gianmaria Frigo, Davide Saladino, Alberto Castagnaro, Francesco Marchiori, et al.
Venue: arXiv
The paper introduces NetLLMeval, a live-network evaluation framework showing that solver architecture and orchestration choices can materially change how well AI systems handle network-administration tasks.

🔗 https://arxiv.org/abs/2606.26960v1

Low Complexity Kolmogorov-Arnold Network-based DPD for Analog RoF Fronthaul

Authors: Carlos Daniel Fontes da Silva, Tianyu Jiang, Lu Zhang, Vjaceslavs Bobrovs, Xianbin Yu, Xiaodan Pang, Oskars Ozolins, Edson Porto da Silva
Venue: arXiv
This work proposes a lower-complexity KAN-based digital predistortion method for analog radio-over-fiber fronthaul, with experimental gains that could matter for scalable high-capacity wireless transport.

🔗 https://arxiv.org/abs/2606.27042v1

MIT/Harvard Events This Week

• June 29 — Group Office Hours: Customer Discovery @ Online
  🔗 https://innovationlabs.harvard.edu/events/group-office-hours-customer-discovery

• June 30 — LL Technology Office Seminar: From 3 Configurations to 300 @ MIT Virtual
  🔗 https://calendar.mit.edu/event/ll-technology-office-seminar-nTop-Bradley-Rothenberg

• June 30 — Summer Seminar Series on Super Pollutants @ MIT Building 55-110
  🔗 https://calendar.mit.edu/event/summer-seminar-series-on-super-pollutants

• July 2 — Founder/Joiner Connections: Pizza + Pitch @ Harvard Innovation Labs
  🔗 https://innovationlabs.harvard.edu/events/founder-joiner-connections-pizza-pitch

Source Issues

• TNT’s calendar page was still stale and mostly surfaced February–April listings, so direct MIT and Harvard pages were used for current events.
• Several rotated X accounts were stale or empty today, especially AST SpaceMobile and OneWeb, so they were skipped instead of padding the digest with weak items.
• Reuters and some vendor press pages were fetch-blocked, so this digest stayed anchored to directly accessible primary sources.

Takeaway

The strongest pattern this morning is controlled deployment: frontier AI is increasingly being packaged with release gates, enterprise rollout structures, and production network operating models instead of shipped as raw capability alone.