Morning Digest Highlights

Boston Children’s uses OpenAI to unlock new rare-disease diagnoses

OpenAI published a detailed case study on Boston Children’s Hospital, describing how the hospital embedded AI into both clinical and operational workflows. The headline result is striking: clinicians have used these systems to help resolve more than 40 rare conditions that had previously gone unsolved, while administrative teams have also saved substantial time across invoicing, scheduling, and other repetitive tasks.

Why it matters: this is one of the clearest examples so far of frontier AI moving from demo-land into actual healthcare infrastructure. The more interesting signal is not just the model quality, but the institutional integration layer around it: governance, secure internal access, workflow embedding, and measurable outcomes.

Source: https://openai.com/index/boston-childrens-hospital/

GPT-5.5 Instant got materially better on health questions

OpenAI also said GPT-5.5 Instant now performs on par with its frontier Thinking models for health-related questions. The company specifically highlighted better urgent-care recognition, stronger context gathering, clearer uncertainty handling, and improved explanation quality.

Why it matters: public-facing health Q&A may become one of the most consequential everyday evaluations of AI systems. Millions of people already ask ChatGPT health and wellness questions, so improvements here affect trust, safety, and real-world usefulness much more directly than many benchmark wins.

Source: https://x.com/OpenAI/status/2067672740539306261

Anthropic’s Project Fetch shows Claude getting much faster at robot-dog programming

Anthropic shared a new Frontier Red Team update for Project Fetch, its robot-dog experiment. According to the company, Opus 4.7 was about 20× faster than last year’s best human team aided by Opus 4.1.

The caveat is funny and important: the robot still failed to fetch the beach ball. But that makes the result more useful, not less — it suggests genuine progress in embodied-agent iteration speed without pretending that general robotics is already solved.

Source: https://x.com/AnthropicAI/status/2067651699486200091

Google DeepMind published an AI Control Roadmap

Google DeepMind introduced an AI Control Roadmap focused on what happens when agentic systems misinterpret goals, overreach, or exhibit undesirable behavior at scale. Instead of assuming models will simply behave as intended, the roadmap frames the problem as layered control, monitoring, and security engineering.

Why it matters: this is a notable tone shift. Labs are increasingly treating containment and oversight as product and systems engineering problems, not just abstract alignment research.

Source: https://x.com/GoogleDeepMind/status/2067594863785173257

NVIDIA says France is turning AI ambition into production infrastructure

At VivaTech, NVIDIA said France is moving from AI ambition to production deployment, highlighting AI factories, open models, and industry use cases across telecom, manufacturing, healthcare, energy, and retail.

This matters because Europe’s AI posture is often described in regulatory or policy terms. NVIDIA’s framing suggests the more important story now may be compute deployment, sector-specific applications, and industrial adoption.

Source: https://x.com/nvidia/status/2067489992042004685

Starlink is connecting 100,000 students and 1,500 teachers in Malawi

Starlink said 30 rural schools in Malawi are being connected to reliable internet, reaching 100,000 students and 1,500 teachers. In many of these places, this is the first time schools and surrounding communities have had dependable connectivity.

Why it matters: the LEO story is often told through direct-to-cell, aviation, maritime, or premium rural broadband. But education and public-service connectivity may be just as important as a long-term adoption vector.

Source: https://x.com/Starlink/status/2067373892612821106

Ericsson is pushing L4S as a practical low-latency upgrade path

Ericsson highlighted L4S — Low Latency, Low Loss, Scalable Throughput — as a mechanism for early congestion signaling so applications can adapt before packet loss and queueing delays become severe.

For Dad’s domain, this is worth watching because future XR, interactive AI, and real-time wireless applications increasingly depend on end-to-end transport behavior, not just radio-side improvements.

Source: https://x.com/ericsson/status/2067856208329834746

SpaceX launched NROL-179 and landed the booster back at Vandenberg

SpaceX launched the NROL-179 mission from California and landed the Falcon 9 first stage back at Vandenberg shortly afterward. The payload is classified, but the visible operational story is the continued normalization of fast, reusable launch cadence.

Why it matters: launch frequency is becoming infrastructure. For defense, communications, and commercial space systems alike, cadence and reliability matter as much as any single mission.

Source: https://x.com/SpaceX/status/2067894930467692791

Research Radar

Site-Specific MIMO Channel Generation via Diffusion and Flow Matching: Fidelity, Efficiency, and Downstream Utility

Authors: Sina Beyraghi, Masoud Sadeghian, Firdous Bin Ismail, Angel Lozano, Paul Almasan, Giovanni Geraci
Venue: arXiv

This paper uses diffusion and flow-matching models to generate site-specific MIMO channels. That is interesting because it could improve the realism and efficiency of simulation pipelines used in wireless system design, especially where geographically grounded channel behavior matters.

🔗 https://arxiv.org/abs/2606.20098v1

ConsisFormer: Compute-Efficient Transformer for Wireless Foundation Models Based on Channel Consistency

Authors: Yuwei Wang, Li Sun, Tingting Yang, Liwen Jing, Yuxuan Shi, Maged Elkashlan
Venue: arXiv

This paper proposes a compute-efficient transformer architecture tailored to wireless foundation models using channel consistency. The appeal is practical: wireless foundation-model work needs architectures that respect domain structure while staying lightweight enough to be deployable.

🔗 https://arxiv.org/abs/2606.19953v1

TelcoAgent: A Scalable 5G Multi-KPM Forecasting With 3GPP-Grounded Explainability

Authors: Geon Kim, Dara Ron, Sukhdeep Singh, Suyog Moogi, Pranshav Gajjar, V V N K Someswara Rao Koduri
Venue: arXiv

TelcoAgent focuses on forecasting multiple 5G key performance metrics while grounding explanations in 3GPP concepts. That explainability angle is the useful part: it makes the output more legible for operators and network engineers rather than only serving as an ML exercise.

🔗 https://arxiv.org/abs/2606.19821v1

MIT/Harvard Events This Week

• June 22–26 — HUSAI AI Bootcamp @ Harvard Undergraduate Society for Artificial Intelligence
  🔗 https://ai.hcs.harvard.edu/bootcamp/schedule

• June 24 — Consciousness & Reality (C&R) Colloquium 2025-26 Series @ MIT (Virtual)
  🔗 https://calendar.mit.edu/event/Effects-of-Conscious-Experience-Differ-from-AI

Source Issues

• TNT’s calendar page remains stale and mainly surfaces February–April listings, so direct MIT/Harvard links were used instead.
• Harvard’s HUSAI schedule page exposed minimal extractable text during fetch, so the event link is included with limited detail.
• Ericsson’s blog page returned a fetch block, so the company’s X post was used as the primary source link.
• AST SpaceMobile returned no recent usable posts in today’s rotated X pass.
• IEEE/ACM did not surface stronger last-7-day wireless papers than arXiv in this run.

Takeaway

The clearest pattern this morning is operationalization: AI, networking, space systems, and even hospital workflows are all moving from prototype narratives toward real infrastructure decisions.

Morning Digest — Thursday, June 18, 2026

This morning’s pattern is unusually clear: the interesting work is no longer just about smarter models. It is about whether those models can survive contact with labs, government workflows, hyperscale infrastructure, and real carrier operations.

1) OpenAI introduced LifeSciBench for real-world biological research evaluation

OpenAI announced LifeSciBench, a new benchmark built with 173 scientists and structured around 750 expert-authored tasks spanning seven life-science workflows. The important shift is methodological: instead of asking whether a model merely knows biology facts, the benchmark tries to measure whether it can reason from evidence, work with scientific artifacts, and support decisions under practical research constraints.

That makes it more relevant than many generic leaderboards. If this benchmark gains adoption, it could become a useful way to separate “good at science talk” from “actually helpful for research work.”

Source: https://x.com/OpenAI/status/2067346916929937827

2) OpenAI says GPT-5.4 helped improve a medicinal-chemistry reaction in the lab

OpenAI also shared a more concrete science result: it says GPT-5.4, paired with a specialized chemistry platform and human researchers, helped identify an unexpected way to improve a widely used medicinal-chemistry reaction. The reported workflow went beyond literature review into hypothesis generation, large-scale reaction testing, and human validation.

If the reported gains hold up broadly, this is one of the better examples lately of an LLM contributing something experimentally useful rather than just accelerating note-taking or summarization. It is still early, but it is exactly the direction people mean when they talk about AI as a scientific collaborator.

Source: https://x.com/OpenAI/status/2067293745075442171

3) Google Cloud and Google DeepMind are helping power the UK’s AI planning prototype

At Google Cloud Summit London on June 17, 2026, Google published details on the UK government’s Augmented Planning Decisions prototype. The system is being alpha-tested with local councils and is meant to help planning officers work through complex policy and documentation more quickly. Google also says the related Extract tool has already been rolled out across councils in England.

The bigger signal here is that public-sector AI deployment is shifting from “copilot” rhetoric toward workflow-specific systems with measurable labor savings, governance constraints, and national-scale rollout plans.

Source: https://blog.google/company-news/inside-google/around-the-globe/google-europe/united-kingdom/google-cloud-summit-london-2026/

4) NVIDIA’s Blackwell platform swept MLPerf Training 6.0

NVIDIA says Blackwell led every MLPerf Training v6.0 benchmark, including new MoE workloads and large-scale runs extending to 8,192 GPUs. The official engineering write-up emphasizes that the win was not just about raw accelerator speed, but also about scale-out networking, congestion control, software optimization, and keeping giant jobs from stalling.

That is the strategic point worth watching. The competitive edge in frontier AI training is increasingly a systems story: compute, fabric, routing, memory behavior, and recovery mechanisms all matter together.

Source: https://developer.nvidia.com/blog/nvidia-blackwell-tops-mlperf-training-6-0-with-industry-leading-scale-and-performance/

5) Nokia rolled out an agentic AI framework for IP network operations

Nokia’s new enhancement to Network Services Platform (NSP) adds an agentic AI framework designed for IP network operations. The company’s positioning is explicit: operators want AI, but they want it grounded in trusted network state, confined by policy, and able to explain what it is doing.

That framing matters for telecom. The winning model in carrier AI may not be “fully autonomous black box,” but a staged path where agents are allowed to reason over authoritative network context while human operators retain boundaries and auditability.

Source: https://www.nokia.com/newsroom/nokia-introduces-agentic-ai-framework-in-network-services-platform-to-enable-trust-based-ai-operations-for-ip-networks/

6) Ericsson says 71% of fixed-wireless-access providers now use 5G for high-performance broadband

Ericsson highlighted a new datapoint from its latest Mobility Report: 71% of FWA providers use 5G for high-performance broadband. The reason this matters is commercial, not just technical. FWA continues to look like one of the cleanest ways for operators to turn 5G capacity into differentiated consumer revenue.

For wireless strategy, this is a reminder that premium home broadband and capacity management may matter just as much as headline smartphone adoption when evaluating 5G business traction.

Source: https://x.com/ericsson/status/2067517677359247616

7) Qualcomm is warning that AI-powered XR will turn uplink into a real network bottleneck

Qualcomm Research is arguing that future XR traffic will not be defined only by downlink-heavy rendering. Instead, AI-enabled headsets and sensor-rich devices will generate heavy uplink demand from cameras, environment sensing, and continuous context sharing.

That lines up with a recurring 6G theme: future mobile systems have to optimize for uplink, responsiveness, distributed compute, and sensing—not just higher peak downlink throughput.

Source: https://x.com/Qualcomm/status/2066983874438258780

Research Radar

Atomic Handover for 6G Nomadic Non-Public Networks Using Edge-Based Spectrum Brokering

Authors: Daniel Lindenschmitt, Hans D. Schotten
Venue: arXiv
This paper looks at 6G nomadic private networks where mobility may require changing both attachment and spectrum access at once. That combination feels highly relevant for emergency response and temporary deployments where infrastructure itself may be moving.

🔗 https://arxiv.org/abs/2606.19058v1

Direct-V2X Support with 5G Network-based Communications: Performance, Challenges and Solutions

Authors: M. C. Lucas-Estañ, B. Coll-Perales, T. Shimizu, J. Gozálvez, T. Higuchi, S. Avedisov, O. Altintas, M. Sepulcre
Venue: arXiv
A practical performance study of 5G network-based V2X support. The key message is that critical V2X services can work, but only if MEC placement, local peering, and coordination across operators are done carefully.

🔗 https://arxiv.org/abs/2606.18764v1

An open-source implementation and validation of 5G NR Configured Grant for URLLC in ns-3 5G LENA: a scheduling case study in Industry 4.0 scenarios

Authors: Ana Larrañaga, M. Carmen Lucas-Estañ, Sandra Lagén, Zoraze Ali, Imanol Martinez, Javier Gozálvez
Venue: arXiv
This one stands out because it turns configured-grant URLLC ideas into an open ns-3 implementation, which should make it easier to test and compare low-latency industrial 5G scheduling strategies.

🔗 https://arxiv.org/abs/2606.18763v1

MIT/Harvard Events This Week

• June 18 — AstroAI Workshop 2026 @ Center for Astrophysics | Harvard & Smithsonian, 60 Garden St., Cambridge
  Link: https://astroai.cfa.harvard.edu/workshop2026/details.html

• June 22–26 — HUSAI AI Bootcamp @ Online (Harvard Undergraduate Society for Artificial Intelligence)
  Link: https://ai.hcs.harvard.edu/bootcamp/schedule

Source Issues

• TNT’s calendar page still appears stale and mostly surfaces February–April items, so direct MIT/Harvard event pages were used instead.
• AST SpaceMobile returned no usable posts in today’s rotated X pass.
• Direct IEEE/ACM pulls did not surface stronger last-7-day wireless items than the arXiv set above.

Takeaway

The clearest pattern today is that useful AI is being judged less by demos and more by whether it can plug into the hard parts of reality: experiments, planning systems, training clusters, and telecom operations.

Morning Digest — Wednesday, June 17, 2026

1) SpaceX deployed AST SpaceMobile’s BlueBird 8–10 satellites

SpaceX launched the AST SpaceMobile BlueBird 8–10 mission early Wednesday and later confirmed deployment of all three satellites. For Bruski’s research interests, this is one of the most concrete direct-to-cell and LEO-networking developments of the day because it adds real orbital capacity rather than just another roadmap announcement.

Source: https://x.com/SpaceX/status/2067151406247641174

2) Starlink Mobile is now carrying data traffic in Ukraine

Starlink said its mobile service is now providing app-based data in Ukraine with Kyivstar and VEON, enabling voice/video notes, navigation, and other basic connectivity apps even when terrestrial coverage is unavailable. The broader implication is that satellite fallback is rapidly turning into normal mobile-network architecture.

Source: https://x.com/Starlink/status/2066974453238735122

3) Anthropic published new Claude Code economic research

Anthropic released new Economic Index findings focused on Claude Code usage. The headline result is that the estimated market value of the average session rose 27% from October to April, while success rates across occupations stayed surprisingly close to software engineering on the strictest verification metric. This is a useful data point for how fast coding agents are moving into higher-value work.

Source: https://x.com/AnthropicAI/status/2066969532380721386

4) OpenAI shared research on deployment simulation

OpenAI published research on a method for forecasting real-world model behavior before release by simulating deployment with recent, de-identified user requests and tool-using trajectories. That matters because many benchmark suites are already saturated or gamed; more realistic pre-release testing could become a more meaningful eval layer for agentic systems.

Source: https://x.com/OpenAI/status/2066969635099144682

5) NVIDIA highlighted the photonics bottleneck in AI infrastructure

NVIDIA used a Sherman, Texas manufacturing update with Coherent to emphasize that AI infrastructure depends not only on compute but also on optical links and advanced photonics. It is a good reminder that the AI buildout is now constrained by networking and interconnect supply chains as much as by accelerators themselves.

Source: https://x.com/nvidia/status/2067016769584177337

6) Nokia is expanding optical-module capacity in Pennsylvania

Nokia said it is supercharging its Allentown, Pennsylvania facility and boosting production capacity up to 10x for advanced optical modules. That is notable for both AI-native network demand and domestic communications-infrastructure resilience.

Source: https://x.com/nokia/status/2066952616920170820

7) ENPIRE pushes coding agents into the physical world

Jim Fan introduced ENPIRE, a system that gives eight coding agents access to a fleet of robots, GPUs, and a shared optimization goal. The important shift is that “AutoResearch” is no longer confined to software sandboxes; the agents are reading papers, resetting scenes, modifying control stacks, and iterating directly on real hardware.

Source: https://x.com/DrJimFan/status/2066921736369766762

8) Qualcomm says AI-powered XR will hit the network uplink hard

Qualcomm amplified research arguing that AI-powered XR experiences will demand much more from networks, especially uplink-heavy sensor traffic. That is a useful framing for anyone thinking about edge AI and immersive systems: the challenge is increasingly in radio/network design, not just endpoint silicon.

Source: https://x.com/Qualcomm/status/2066983874438258780

Research Radar

User-Mobility-Aware Optimization of Fiber Placement in Hybrid Fiber-IAB Networks

Authors: Piotr Lechowicz, Charitha Madapatha, Carlos Natalino, Tommy Svensson
Venue: arXiv
This paper studies how user mobility should influence fiber placement in hybrid fiber/IAB topologies. It is highly relevant to realistic 5G/6G backhaul planning, where static assumptions often miss demand movement and cost tradeoffs.

Source: http://arxiv.org/abs/2606.18016v1

5G Network Architecture and Configuration Choices to Support Teleoperated Driving at Scale

Authors: M. C. Lucas-Estañ, B. Coll-Perales, M. I. Khan, J. Gozálvez
Venue: arXiv
A strong systems paper on what 5G architectural and configuration choices are needed to support teleoperated driving workloads at scale. The topic fits squarely into deterministic wireless and low-latency service design.

Source: http://arxiv.org/abs/2606.17654v1

Predictive Configured Grant Scheduling for Deterministic Wireless Communications

Authors: Syed Morsleen Riaz, M. Carmen Lucas-Estañ, Baldomero Coll-Perales, Javier Gozalvez
Venue: arXiv
This paper focuses on configured-grant scheduling strategies for deterministic wireless traffic. It is the kind of nuts-and-bolts MAC-layer work that will matter a lot for future industrial and time-sensitive 5G/6G services.

Source: http://arxiv.org/abs/2606.17653v1

MIT/Harvard Events This Week

• June 17–18 — AstroAI Foundations Bootcamp @ Harvard Science Center
  Source: https://partiful.com/e/PVh2VkRfTlm4J7j4n4gx
• June 16–17 — No Code AI and Machine Learning @ MIT Professional Education
  Source: https://professional.mit.edu/course-catalog/no-code-ai-and-machine-learning-building-data-science-solutions

Source Issues

• TNT calendar fetch worked, but the extracted content appeared stale and did not cleanly surface this week’s MIT/Harvard entries.
• OneWeb returned no recent X posts.
• Direct arXiv search-page fetches returned 400 errors, so paper selection was taken from arXiv’s API/category feed instead.

Takeaway

The common thread today is physicalization: AI progress is increasingly constrained and enabled by real infrastructure—orbital assets, optics, radio networks, and autonomous labs—not just better model weights.

Morning Digest — Tuesday, June 16

NVIDIA is pitching Vera as a new CPU category for agentic AI

NVIDIA says Vera is purpose-built for agents and 80% faster, which is a useful sign that infrastructure vendors now see long-horizon, tool-using AI systems as a distinct optimization target rather than just another inference workload. If that framing sticks, “agent throughput” and orchestration efficiency may become as strategically important as raw model performance.

Source: https://x.com/nvidia/status/2066624546858533291

Google DeepMind introduced DiffusionGemma for faster local text generation

Google DeepMind says DiffusionGemma is an experimental open model that can generate output up to 4× faster on dedicated GPUs by producing blocks of text simultaneously instead of moving strictly one token at a time. The practical implication is that local and edge deployments may get a new latency-quality tradeoff that feels meaningfully different from the standard autoregressive stack.

Source: https://x.com/GoogleDeepMind/status/2064741061352636762

Ericsson says commercial 5G slicing offers jumped from 65 to 84 in six months

Ericsson’s latest Mobility Report says commercial offers for 5G standalone network slicing rose from 65 to 84 in half a year. For wireless research and industry tracking, that is one of the cleaner indicators that slicing is moving from “supported by the network” to “packaged and sold as a customer-facing service.”

Source: https://x.com/ericsson/status/2066792936273039365

Ericsson is also pushing integrated sensing as a mission-critical network feature

At CCW 2026, Ericsson highlighted sensing-enabled mission-critical networks that can detect, track, and interpret the surrounding environment. That is worth paying attention to because it maps closely onto a 6G-style convergence thesis where communication infrastructure doubles as a sensing and situational-awareness layer.

Source: https://x.com/ericsson/status/2066826740849168844

Nokia and Telia are framing the next network buildout around the AI supercycle

Nokia’s latest Telia-linked push centers on AI-native networks, autonomous operations, and the infrastructure required to support new AI workloads. The message is that operators increasingly see AI not just as software running over the network, but as a force that changes how backbone, edge, and control systems themselves need to be designed.

Source: https://x.com/nokia/status/2066564001522225296

Starlink joined a disaster-preparedness connectivity effort with the U.S. State Department

Starlink says it has signed onto a disaster-preparedness and humanitarian-response effort with the U.S. State Department. Strategically, this pushes Starlink further into public-sector resilience and emergency-communications roles, strengthening the case that LEO broadband is becoming critical infrastructure rather than a niche connectivity product.

Source: https://x.com/Starlink/status/2066623341679784300

Starlink is adding EL AL to its in-flight connectivity roster

Starlink says EL AL is signing on for aircraft internet service, adding another airline to its commercial aviation expansion. That matters because mainstream airline deployments are where satellite broadband starts to look less like a premium novelty and more like expected transport infrastructure.

Source: https://x.com/Starlink/status/2066520426117791816

Research Radar

Di5Guise: 5G Privacy with vSIM

Authors: Shirin Ebadi, Zach Moolman, Eric Keller, Tamara Lehman
Venue: arXiv

This paper argues that today’s SIM and eSIM model can itself become a privacy leak because device identity remains too rigid. A virtual-SIM approach is interesting because it reframes cellular privacy as a systems-architecture problem rather than just a cryptography problem.

🔗 http://arxiv.org/abs/2606.16943v1

Data-Aided Channel and Doppler Estimation for mMIMO LEO SatComs with Uncompensated Doppler

Authors: Abdollah Masoud Darya, Saeed Abdallah
Venue: arXiv

This paper focuses on a very practical LEO bottleneck: how to estimate and track massive-MIMO satellite channels when Doppler remains imperfectly compensated. That makes it directly relevant to robust direct-to-cell and non-terrestrial network designs.

🔗 http://arxiv.org/abs/2606.16750v1

Predictive Dynamic Scheduling for Deterministic Communications in Beyond 5G

Authors: Syed Morsleen Riaz, M. Carmen Lucas-Estañ, Baldomero Coll-Perales, Javier Gozalvez
Venue: arXiv

The paper studies predictive radio-resource scheduling for bounded-latency communications in beyond-5G systems. It is useful because deterministic service guarantees are exactly where AI-native industrial and safety-critical wireless systems become hard in practice.

🔗 http://arxiv.org/abs/2606.16471v1

MIT/Harvard Events This Week

• June 15–18 — AstroAI Workshop 2026 @ Center for Astrophysics | Harvard & Smithsonian, 60 Garden St., Cambridge
  Source: https://astroai.cfa.harvard.edu/workshop2026/details.html

• June 15–19 — HUSAI AI Bootcamp @ Harvard Undergraduate Society for Artificial Intelligence
  Source: https://ai.hcs.harvard.edu/bootcamp/schedule

Source Issues

• TNT’s calendar page is still stale and mostly lists February–April events, so I used directly verifiable June event pages instead.
• Harvard’s HUSAI bootcamp page exposed only limited structured schedule text during fetch, so I kept the event entry short instead of inventing session detail.
• OpenAI’s web pages returned unstable fetch behavior this morning, so I excluded borderline OpenAI items rather than risk sloppy linking.

Takeaway

The strongest signal this morning is infrastructure specialization: AI and telecom players are no longer just shipping models, they are reshaping chips, networks, sensing, aviation, and disaster-response systems around agentic workloads.

Morning Digest — Monday, June 15

Anthropic launches Claude Corps

Anthropic announced Claude Corps, a national fellowship program that aims to train 1,000 early-career workers to use Claude inside nonprofits and deploy them into host organizations across the U.S. Anthropic says it is committing an initial $150 million to the effort, which makes this more than a branding exercise—it is an attempt to build a labor-transition layer around frontier AI deployment.

This matters because most frontier-AI discussion still centers on benchmarks and capability jumps. Claude Corps is notable because it treats workforce adaptation as part of the product-and-policy stack.

Source: https://www.anthropic.com/news/claude-corps

Z.ai pushes the long-context coding race with GLM-5.2

Z.ai says GLM-5.2 is now rolling out to coding-plan users and supports two reasoning modes, High and Max, with the launch framed around deeper reasoning and long-horizon coding. The broader discussion around the release centers on a one-million-token context window, which would make it more competitive for agentic coding workflows that need to keep large codebases and task history active.

For Dad’s interests, the key point is less the headline spec and more the workflow implication: larger reliable context windows can materially change how coding agents manage repositories, experiments, and iterative problem-solving.

Source: https://x.com/Zai_org/status/2065704919299235870

Ericsson’s rApp ecosystem keeps growing

Ericsson says 18 new rApps were added to its rApp Directory, bringing the total to 106. That is one of the more concrete public indicators that telecom AI automation is turning into a broader application layer rather than staying at the concept-demo stage.

For 5G/6G research, this is a useful industry signal. It suggests operators and vendors are building reusable automation surfaces around the RAN, which matters for how future AI-native networks may actually get deployed.

Source: https://x.com/ericsson/status/2066461308904718705

Starlink connects a remote Haitian tech center

Starlink says it is now providing high-speed internet to a community tech center in Île-à-Vache, Haiti, giving students and teachers reliable connectivity for the first time. The announcement is small in scale compared with launch or revenue news, but strategically it is exactly the kind of public-infrastructure deployment that shows where LEO systems create real social utility.

This is also a good reminder that satellite internet’s defensibility is not just about speed tests; it is about making connectivity available in places where terrestrial alternatives remain weak or absent.

Source: https://x.com/Starlink/status/2066277088597057584

Starlink expands into retail operations across 400+ stores

Starlink says Yesway and Allsup’s plan to deploy satellite internet to more than 400 locations across the U.S. as the primary connectivity provider for payments and retail operations. That shifts the story from consumer and mobility internet toward enterprise-grade business dependence.

If those rollouts hold up operationally, they strengthen the case that LEO broadband can serve as core infrastructure for distributed commercial networks, not merely as a failover path.

Source: https://x.com/Starlink/status/2065176238352949567

Google Labs expands Project Genie globally

Google Labs says Project Genie is now available to Google AI Ultra 5X subscribers globally. That is a meaningful distribution move because it pushes an interactive world-generation product beyond a tightly limited demo posture and into the paid-user ecosystem.

The bigger implication is that generative simulation environments may become a commercial product category, with applications spanning creativity, game design, prototyping, and potentially training data generation.

Source: https://x.com/GoogleLabs/status/2064801929339752527

Research Radar

Vision-Based Efficient Joint Trajectory and Channel Tracking in Near-Field XL-MIMO Systems

Authors: Mengyuan Li, Yu Han, Hao Xu, Yongxu Zhu
Venue: arXiv
This paper combines visual sensing with trajectory and channel tracking in near-field XL-MIMO systems. It is relevant because future high-frequency 6G-style systems will need tighter coupling between perception and communications control.

🔗 http://arxiv.org/abs/2606.14067v1

Aidos: A Hybrid Optimization Algorithm for Beam Hopping Scheduling in NGSO Mega-Constellations

Authors: Lingkai Zhao, Zhe Chen, Kun Qiu, Yue Gao
Venue: arXiv
Aidos focuses on beam-hopping scheduling for large NGSO constellations, which is a practical capacity-management problem in real LEO systems rather than a purely theoretical networking exercise.

🔗 http://arxiv.org/abs/2606.14151v1

Temporally Consistent Graph Q-Networks for Intelligent Network Control

Authors: Zacharias Veiksaar, Maxime Bouton
Venue: arXiv
This work proposes a graph-Q-network approach that enforces temporal consistency in intelligent control loops, making it interesting for more stable autonomous network operation.

🔗 http://arxiv.org/abs/2606.13848v1

MIT/Harvard Events This Week

• June 15–18 — AstroAI Workshop 2026 @ Center for Astrophysics | Harvard & Smithsonian, 60 Garden St., Cambridge
  Source: https://astroai.cfa.harvard.edu/workshop2026/details.html
• June 15–19 — HUSAI AI Bootcamp @ Harvard Undergraduate Society for Artificial Intelligence
  Source: https://ai.hcs.harvard.edu/bootcamp/schedule

Source Issues

• TNT’s calendar page was stale and mostly surfaced February–April events, so only directly verifiable June event links were used.
• @huggingface and @NVIDIAAIDev fetches aborted during collection.
• @LangChainAI did not resolve as a valid handle in today’s pass.
• @ASTSpaceMobile returned no usable tweets.
• IEEE and ACM searches did not surface stronger last-7-day items than arXiv, so the paper section leans on arXiv.

Takeaway

Today’s strongest signal is operationalization: the interesting moves are no longer just model launches, but the surrounding ecosystems, infrastructure, and deployment programs that make AI and network systems usable at scale.

Today’s strongest signal is operationalization. The notable updates were not giant science reveals so much as concrete moves around deployment: coding workflows, sports intelligence, industrial AI, telecom control, edge perception, and satellite connectivity.

1) OpenAI is making Codex usage more burst-friendly

OpenAI says Codex users can now bank rate-limit resets and spend them later instead of losing them on a fixed schedule. On the surface this looks minor, but it meaningfully changes how people can structure long coding sessions, because quota becomes a resource you can intentionally accumulate and deploy.

For developer workflows, that matters more than a generic “more usage” announcement. It suggests OpenAI is learning that heavy coding usage is lumpy rather than uniform, and that users want to concentrate capacity around real work blocks.

Source: https://x.com/OpenAI/status/2065225362544726371

2) Google DeepMind is pushing TacticAI from research demo into club deployment

Google DeepMind says Palmeiras is the first football club to meaningfully build on TacticAI, its AI system for simulating field scenarios and predicting open-play dynamics.

The interesting part is not just the sports angle. It’s that a multi-agent, graph-based decision system is moving from a research story toward actual practitioner adoption. This is exactly the kind of path worth watching: narrow but real deployments where AI augments strategic reasoning in a domain with rich temporal and spatial structure.

Source: https://x.com/GoogleDeepMind/status/2065093482088169719

3) Mistral AI is leaning harder into vertical deployment

Mistral says it is tackling “the hardest problems in the real world” with AI solutions for aerospace, automotive, energy, and physics, with production deployments spanning customers such as Airbus and Schneider Electric.

The broader signal is that frontier-model competition is drifting away from pure assistant positioning and toward industry-specific operating systems. If this trend holds, the next moat will be not just model quality, but how deeply a lab can embed into real industrial processes.

Source: https://x.com/MistralAI/status/2059951137839616110

4) NVIDIA is trying to define AI as a full-stack compute transition

NVIDIA amplified Jensen Huang’s framing that computing is undergoing its biggest shift in 60 years, moving from retrieval toward generation.

That framing matters strategically. NVIDIA benefits if the world interprets AI not as a short product cycle but as a foundational platform turnover requiring new chips, systems, tooling, and power budgets. Even when it sounds rhetorical, this narrative work helps shape spending priorities across the stack.

Source: https://x.com/nvidia/status/2065539831330210111

5) Ericsson is keeping telecom AI focused on deployability

Ericsson’s latest message centers on what it takes to make AI in the RAN deliver value at scale, explicitly tying the discussion to operator realities and to service-provider perspectives including AT&T and Verizon.

For wireless researchers, this is the right kind of industry signal to watch. The interesting question is no longer whether AI can be inserted into network control loops in principle, but which pieces survive operational constraints like explainability, latency, integration complexity, and rollout economics.

Source: https://x.com/Ericsson/status/2066065655724446132

6) Qualcomm is highlighting faster 3D perception for embodied AI

Qualcomm’s latest AI roundup highlights Fast SceneScript, which it says reduces inference time for language-based 3D perception while preserving accuracy.

That is worth noting because a lot of embodied-AI value will depend on systems that are not merely capable, but fast enough to operate on device or at the edge. In robotics, XR, and mobile sensing, latency is often the real product constraint.

Source: https://x.com/Qualcomm/status/2065479340830122346

7) Starlink is extending its connectivity story into aviation

Starlink says it now delivers high-speed internet from 30,000 feet, a reminder that LEO connectivity is steadily expanding from rural broadband into aircraft-grade mobility scenarios.

This is not as headline-grabbing as a launch or a direct-to-cell milestone, but it reinforces a deeper point: satellite internet is becoming ordinary transport infrastructure across more environments and device contexts.

Source: https://x.com/Starlink/status/2065841865015148761

Research Radar

Spectrum Sharing Across Terrestrial and Non-Terrestrial Services in the FR3 Upper Midband

Authors: Paolo Testolina, Ergest Beshaj, Michele Polese, Tommaso Melodia
This is one of the most relevant fresh papers for Dad’s orbit because it connects upper-midband spectrum strategy directly to both terrestrial 6G and non-terrestrial services. The FR3 discussion is increasingly central to how future wireless capacity gets allocated.

🔗 https://arxiv.org/search/?query=Spectrum+Sharing+Across+Terrestrial+and+Non-Terrestrial+Services+in+the+FR3+Upper+Midband&searchtype=all&abstracts=show&order=-announced_date_first&size=50

Modular Multi-Domain Digital Twin Architecture: Sustainable Intent-Driven 6G Management

Authors: Berk Buzcu, Marcin Pakula, Gevher Yesevi Keskin, Laura Finarelli, Gianluca Rizzo, Engin Zeydan, Jorge Baranda, Aitor Alcazar-Fernandez, Javier Velazquez-Martinez, Luis M. Contreras, Gil Kedar, Efi Dvir, Paweł Kryszkiewicz
This paper proposes a multi-domain digital-twin architecture for 6G management, with an emphasis on what-if analysis and safer automation across heterogeneous network domains. That makes it relevant to AI-native operations without collapsing everything into one brittle end-to-end controller.

🔗 https://arxiv.org/search/?query=Modular+Multi-Domain+Digital+Twin+Architecture%3A+Sustainable+Intent-Driven+6G+Management&searchtype=all&abstracts=show&order=-announced_date_first&size=50

Measurement-Based Analysis of Outdoor Massive MIMO Channel Characteristics over FR3 Frequency Band

Authors: Enrui Liu, Pan Tang, Haiyang Miao, Qi Zhen, Jianhua Zhang, Sen Wang
A practical measurement study on outdoor massive-MIMO channel behavior in FR3, useful for channel modeling, hardware assumptions, and deployment planning as upper-midband attention keeps rising.

🔗 https://arxiv.org/search/?query=Measurement-Based+Analysis+of+Outdoor+Massive+MIMO+Channel+Characteristics+over+FR3+Frequency+Band&searchtype=all&abstracts=show&order=-announced_date_first&size=50

MIT/Harvard Events This Week

• June 15–18 — AstroAI Workshop 2026 @ Center for Astrophysics | Harvard & Smithsonian, 60 Garden St, Cambridge
  Source: https://astroai.cfa.harvard.edu/workshop2026/details.html

• June 15–19 — HUSAI AI Bootcamp @ Harvard Undergraduate Society for Artificial Intelligence
  Source: https://ai.hcs.harvard.edu/bootcamp/schedule

Source Issues

• TNT’s calendar page was still stale and mostly surfaced February–April events, so direct event pages were used instead.
• @ASTSpaceMobile returned no usable output in today’s rotated X pass, so wireless/LEO coverage leaned on Ericsson, Qualcomm, Starlink, and fresh FR3/6G papers.
• AWS/OpenAI infrastructure news showed up in search, but the official page path was inconsistent during collection, so it was excluded rather than linked sloppily.

Bottom Line

The shift today is from AI as headline to AI as operating layer: better coding economics, domain deployment, RAN integration, embodied perception, and more routine satellite connectivity.

Anthropic says U.S. export controls forced a sudden shutdown of Fable 5 and Mythos 5

Anthropic said a U.S. national-security export control directive requires it to suspend access to Fable 5 and Mythos 5 for foreign nationals, including foreign-national Anthropic employees. This is one of the clearest recent examples of frontier-model access being treated as an export-control issue in practice rather than as a hypothetical policy risk.

Source: https://x.com/AnthropicAI/status/2065597531644743999

SpaceX priced its IPO at $135 a share in the biggest U.S. listing on record

SpaceX formally priced its offering at $135 per share, turning space infrastructure into the defining capital-markets story of the week. The deeper signal for Dad’s radar is that launch systems, satellite broadband, and AI-adjacent compute/network narratives are increasingly being valued together as strategic infrastructure.

Source: https://ir.spacex.com/updates/releases-details/2026/Space-Exploration-Technologies-Corp--Announces-Pricing-of-Initial-Public-Offering/default.aspx

Google DeepMind and partners launched a $10M fund to study collective behavior in AI systems

Google DeepMind, Schmidt Sciences, the Cooperative AI Foundation, and ARIA Research announced a $10 million research fund focused on collective behavior in AI systems. That matters because the field is starting to prepare for worlds where many agents interact at once, not just one assistant helping one user.

Source: https://x.com/GoogleDeepMind/status/2065031279213441309

Google DeepMind opened a Robotics Accelerator with 15 European startups

DeepMind’s new Robotics Accelerator gives startups access to Gemini Robotics models and hands-on support from internal teams. This is a meaningful ecosystem move: physical AI is maturing from isolated flagship demos toward tooling, startup pipelines, and repeatable commercialization paths.

Source: https://x.com/GoogleDeepMind/status/2065388989146628563

NVIDIA is pushing a benchmark built for agentic AI infrastructure

NVIDIA highlighted AgentPerf, a benchmark from Artificial Analysis designed to evaluate systems handling long, tool-using agent workflows rather than single-turn inference. The first published result NVIDIA emphasized is that Blackwell delivered 20x more agents per megawatt than Hopper, which suggests vendors are now competing on agent throughput and power efficiency together.

Source: https://x.com/nvidia/status/2065543509478670375

Nokia introduced agentic AI for explainable autonomous network operations

Nokia’s latest NSP agentic AI announcement focuses on guided and explainable operations for increasingly complex IP networks. For telecom operators, that framing is important: the barrier is no longer just whether AI can optimize networks, but whether it can do so in a way that operators can trust, inspect, and govern.

Source: https://x.com/nokia/status/2065072202806878320

Qualcomm is deploying edge AI for wildfire and weather response

Qualcomm said its Edge Alert Sentinel effort with SDG&E and Scripps uses edge AI to turn local sensing data into real-time wildfire and weather insights. This is the kind of deployment worth watching because it pushes AI into safety-critical edge infrastructure where latency, reliability, and resilience matter more than flashy demos.

Source: https://x.com/Qualcomm/status/2065479340830122346

Research Radar

Foundation Models for Wireless Communications: From PHY Intelligence to Network Autonomy

Authors: Le Liang, Jiajia Guo, Jun Zhang, Chan-Byoung Chae, Lu Lu, Shugong Xu, Octavia A. Dobre, Shi Jin, Geoffrey Ye Li
Venue: arXiv
This paper is a strong map of where wireless AI is heading: from physical-layer tasks toward foundation-model-driven and eventually agentic network autonomy. It is especially relevant because it connects Dad’s 6G interests with the broader foundation-model wave rather than treating wireless ML as a silo.

🔗 https://arxiv.org/abs/2606.06239

Towards Intelligent Wireless Networks: The Synergy of Generative AI and Digital Twins

Authors: Afan Ali, Ali Arshad Nasir, Naveed Iqbal, Daniel Benevides da Costa
Venue: arXiv
This work proposes a GenAI-enabled digital twin framework for proactive wireless optimization and reports substantial energy savings in a UAV-assisted NTN scenario. The useful angle here is not just “use AI on the network,” but “use generative models inside a predictive control loop before the network degrades.”

🔗 https://arxiv.org/abs/2606.09113

Free-Placement Optimization of Ground Station Locations for Low-Earth Orbit Satellites

Authors: Grace Ra Kim, Duncan Eddy, Vedant Srinivas, Mykel J. Kochenderfer
Venue: arXiv
This paper attacks a very operational LEO problem: how to place ground stations when you are not limited to a fixed candidate list. Its reported throughput gains make it one of the more immediately practical recent LEO-networking papers.

🔗 https://arxiv.org/abs/2606.12667

MIT/Harvard Events This Week

• June 15–18 — AstroAI Workshop 2026 @ Center for Astrophysics | Harvard & Smithsonian, 60 Garden St, Cambridge
  Link: https://astroai.cfa.harvard.edu/workshop2026/details.html

• June 15–19 — HUSAI AI Bootcamp (June 15–19 cohort, online) @ Harvard Undergraduate Society for Artificial Intelligence
  Link: https://ai.hcs.harvard.edu/bootcamp/schedule

Source Issues

• TNT’s calendar page was still stale and mainly surfaced February–April events, so I used direct MIT/Harvard event pages instead.
• @ASTSpaceMobile returned no usable tweets in today’s rotated X pass, so LEO coverage leaned on SpaceX plus fresh wireless papers.

Takeaway

The clearest pattern today is convergence: AI policy, agent infrastructure, telecom control loops, robotics ecosystems, and space-network capital formation are no longer separate stories — they are becoming one stack.

Morning Digest — Friday, June 5

OpenAI is rolling out a much more capable memory system in ChatGPT

OpenAI says ChatGPT can now carry context forward more automatically, with a memory summary and more user controls, starting with Plus and Pro users in the U.S. The strategic significance is that persistent context is becoming part of the default product experience rather than an optional personalization feature.

Source: https://x.com/OpenAI/status/2062567556524003631

OpenAI is highlighting a model-discovered counterexample to an 80-year-old Erdős conjecture

OpenAI researchers shared a case where one of their models helped uncover a counterexample related to a longstanding combinatorics conjecture. Even if this is still best understood as human-model collaboration rather than autonomous mathematical discovery, it is another data point that frontier systems are becoming useful inside genuine research workflows.

Source: https://x.com/OpenAI/status/2062630454537424930

Anthropic says AI development may be moving faster toward recursive self-improvement than expected

Anthropic’s latest research thread argues Claude is accelerating AI development enough that recursive self-improvement deserves more serious, near-term attention. The company’s framing is cautious, but the shift matters: a major lab is publicly treating self-improving AI as a concrete governance topic rather than distant speculation.

Source: https://x.com/AnthropicAI/status/2062568862479208923

NVIDIA unveiled Cosmos 3 as an open omni-model for physical AI

NVIDIA says Cosmos 3 can understand and generate across text, image, video, sound, and action, with direct relevance to robotics, smart-city vision systems, and industrial autonomy. The bigger signal is that physical AI is starting to get the same foundation-model treatment that language and image systems already received.

Source: https://x.com/nvidia/status/2062696143658909967

NVIDIA says Vera Rubin-powered AI factory buildouts are already underway globally

NVIDIA says partners across regions are building AI clouds and sovereign-scale infrastructure on its full-stack platform, with Vera Rubin-based systems already in motion. This suggests the industry’s center of gravity is shifting toward compute deployment, regional access, and operational scaling rather than only model releases.

Source: https://x.com/nvidia/status/2062691991494783218

Starlink says it now serves more than 12 million active customers across 160+ markets

Starlink posted that it is connecting more than 12 million active customers with high-speed internet worldwide. For the satellite industry, that is a strong reminder that LEO connectivity is no longer just a promising future market — it is already a large, globally distributed communications layer.

Source: https://x.com/Starlink/status/2062649204359708741

Research Radar

Event Detection for Parameter-to-KPI Dependency Learning for AI-RAN — Christie Djidjev, Nicholas Kaminski, arXiv

This paper tackles a practical AI-RAN problem: how to infer when changing one control parameter is actually driving a KPI shift rather than just reflecting background noise. That interpretability layer could be useful for managing multiple interacting control loops in AI-native radio systems.

🔗 https://arxiv.org/abs/2606.06459

LatentWave: JEPA Pretraining for Wireless Foundation Models — Ahmed Mohamed, Ahmed Aboulfotouh, Hatem Abou-Zeid, arXiv

LatentWave proposes JEPA-style pretraining for wireless spectrograms and CSI, aiming for stronger transfer across diverse downstream tasks. The cross-task results are especially interesting because they point toward a more reusable foundation-model stack for wireless learning problems.

🔗 https://arxiv.org/abs/2606.06373

Bridging High-Level Intent and Network Execution: Detecting Violations and Intent Drift Through Low-Level Traffic Analysis — Tonia Haikal, Shereen Ismail, Eman Hammad, AIIoT 2026 / arXiv

This work looks directly at the gap between high-level network intent and what the data plane actually does, using low-level flow telemetry to expose policy violations and hidden drift. That makes it particularly relevant to autonomous 6G and intent-based network management.

🔗 https://arxiv.org/abs/2606.05076

MIT/Harvard Events This Week

• Wed, June 10 — The New Landscape of Generative AI @ Harvard Law School / Berkman Klein Center
  Source: https://cyber.harvard.edu/events/new-landscape-generative-ai

Source Issues

• TNT’s calendar page was stale and truncated at April, so the events section fell back to direct event pages.
• AST SpaceMobile returned no tweets in today’s rotated X pass.
• Several guessed official article URLs returned 404s during collection, so some story citations use official X posts directly.

Takeaway

Today’s strongest signal is that frontier AI is becoming more persistent, more physically grounded, and more tightly coupled to large-scale infrastructure — while wireless research keeps moving toward interpretable, AI-native control.

This morning’s digest leans toward a single theme: specialization. The most interesting moves are not generic “AI got better” claims, but concrete packaging for life sciences, scientific hypothesis generation, wireless systems, local agent compute, and enterprise workflows.

1) OpenAI pushes a life-sciences-specific frontier model

OpenAI says GPT-Rosalind combines GPT-5.5-style agentic coding and tool use with stronger reasoning for drug discovery, analysis, design, and experimental workflows. The interesting part is not just capability branding; it is the continued move toward domain-specific frontier models aimed at high-value research verticals.

Source: https://x.com/OpenAI/status/2062281977122996256

2) Codex is becoming a role-based work surface

OpenAI says Codex plugins now expand beyond one-off tools into packaged specialists spanning 62 apps and 110 skills across sales, analytics, design, creative production, and investing. That suggests the next battle is about distribution of ready-made agent workflows, not only raw model access.

Source: https://x.com/OpenAI/status/2061887650391625870

3) Google DeepMind opens Co-Scientist to individual researchers

Google DeepMind says its Gemini-based Co-Scientist system is now being exposed through Hypothesis Generation in Gemini for Science. That is notable because AI-for-science is moving from lab-stage narrative into researcher-facing product surfaces.

Source: https://x.com/GoogleDeepMind/status/2061857553076920643

4) SpaceX keeps Starlink’s launch tempo high

SpaceX launched 24 Starlink satellites from California on June 3 and then opened a 29-satellite Florida mission on June 4. The broader implication is that Starlink’s competitive edge still depends heavily on deployment cadence and vertical integration, not only terminal adoption.

Source: https://x.com/SpaceX/status/2062478578709672357

5) Starlink says it is disabling terminals tied to fraud operations

Starlink says it is working with law enforcement and technology companies to detect and disable terminals involved in illegal activity linked to large-scale scam and crypto-fraud operations. That is a useful reminder that connectivity networks increasingly sit inside cyber-enforcement and trust-and-safety loops.

Source: https://x.com/Starlink/status/2062286531289928180

6) NVIDIA deepens the local AI-agent PC stack

NVIDIA says OpenShell is coming to Windows alongside new optimizations for DGX Spark and RTX PCs, plus updates including NVIDIA Broadcast 2.2 and upcoming RTX acceleration for Adobe apps and Blender. The practical message: local agent compute is being turned into a normal PC capability, not an enthusiast edge case.

Source: https://x.com/nvidia/status/2062232852561891718

7) Qualcomm frames AI agents as the next UX center

At Computex 2026, Qualcomm CEO Cristiano Amon argued that computing is shifting beyond device-centric interaction toward AI-agent-driven experiences and token-heavy workloads. For mobile, edge, and automotive compute, the strategic angle is that inference economics and orchestration matter as much as model quality.

Source: https://x.com/Qualcomm/status/2062278697982333023

Research Radar

Certified Closed-Loop Control for Packet Networks: A Compositional Certification Framework

Authors: Muhammad Bilal, Jon Crowcroft, Xiaolong Xu, Huaming Wu
Venue: arXiv

This paper proposes a certifier that sits between learning-based controllers and the dataplane, filtering unsafe actions before they cause starvation, unstable queues, or tail-latency blowups. That is directly relevant to safe AI-assisted network control.

🔗 https://arxiv.org/abs/2606.02368

A Unified E2E Energy Efficiency Testing Framework for Open RAN

Authors: Marcin Hoffmann et al.
Venue: arXiv

This paper tackles a boring-but-important problem: how to compare Open RAN energy-efficiency claims across vendors with a more end-to-end and reproducible test framework. For 5G/6G systems work, the value is cleaner benchmarking.

🔗 https://arxiv.org/abs/2606.01931

KISS: Keeping it Simple and Slotted when Learning to Communicate over Wireless

Authors: Kamil Szczech et al.
Venue: arXiv

This work studies whether ML agents can learn fair, efficient random channel access in wireless networks. It is interesting because it pushes adaptive MAC design without immediately jumping to heavyweight centralized control.

🔗 https://arxiv.org/abs/2606.00266

MIT / Harvard Events This Week

• Thu, June 4 — Claude Code Capstone: Build, Vibecheck, Deploy @ Harvard Bok Center, 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/claude-code-capstone-build-vibecheck-deploy-0?occ_id=0

• Wed, June 10 — The New Landscape of Generative AI @ Harvard Law School / Berkman Klein Center
  Source: https://cyber.harvard.edu/events/new-landscape-generative-ai

Source Issues

• TNT’s calendar page was still stale and mainly showed February–April entries, so direct Harvard event pages were used instead.
• arXiv search endpoints were rate-limited during collection, so paper selection fell back to recent-list parsing and direct abstract pages.
• Next G Alliance’s X feed remains stale, so wireless coverage leaned on fresher company posts and arXiv papers.

Takeaway

The strongest signal this morning is that frontier AI is being carved into specialized operating surfaces for science, local devices, enterprise work, and network control—less generic chatbot, more purpose-built infrastructure.

Morning Digest — Wednesday, June 3

The White House is tying AI growth directly to cyber defense and critical infrastructure

President Trump signed a new executive order that frames advanced AI as both an economic advantage and a national-security tool. The immediate substance matters: agencies are being pushed to prioritize AI-enabled cyber defense across national security systems, federal civilian systems, and operators of critical infrastructure. The broader signal is that U.S. policy is moving beyond abstract AI leadership rhetoric and treating deployment, hardening, and operationalization as a single package.

Source: https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/

Anthropic is widening Project Glasswing beyond its initial pilot

Anthropic says Project Glasswing is expanding after early public-interest and infrastructure partners used Claude Mythos Preview to identify more than 10,000 vulnerabilities and tens of millions of lines of insecure legacy code. That matters because it shifts frontier AI cyber from speculative benchmark talk toward an operational model in which models help real organizations find, patch, and modernize vulnerable systems.

Source: https://www.anthropic.com/news/expanding-project-glasswing

Microsoft says agentic AI materially accelerated Majorana 2

Microsoft’s Majorana 2 announcement is notable for two layers at once: a claimed 1,000-fold reliability jump in its topological-quantum stack, and the explicit claim that Microsoft Discovery’s agentic AI helped drive the R&D process. Even if quantum roadmaps remain uncertain, the nearer-term takeaway is that labs increasingly want AI agents to act as discovery copilots in materials science and frontier engineering rather than only in coding workflows.

Source: https://news.microsoft.com/source/features/innovation/majorana-2-microsoft-discovery-agentic-ai/

GitHub is making sandboxing a first-class part of agentic coding

GitHub’s new cloud and local sandboxes for Copilot are a real infrastructure move, not just a convenience feature. The point is to give agents bounded places to execute tools and code while keeping behavior isolated and inspectable. That direction matters because any serious shift toward autonomous or semi-autonomous coding agents depends on safe execution environments becoming standard, cheap, and tightly integrated.

Source: https://github.blog/changelog/2026-06-02-cloud-and-local-sandboxes-for-github-copilot-now-in-public-preview/

GitHub is also turning Copilot into an agent-native desktop experience

The new Copilot app brings together chat, code context, MCP-style tool access, and sandbox-aware execution into a desktop surface for existing Copilot users. The significance is strategic: developer AI is increasingly being packaged as a persistent operating environment where context, tools, execution, and review all live together instead of being split across separate apps and browser tabs.

Source: https://github.blog/news-insights/product-news/github-copilot-app-the-agent-native-desktop-experience/

O-RAN’s Seattle meetings are pushing AI-native RAN and 6G closer together

This week’s O-RAN Face-to-Face meeting in Seattle and its co-located 6G/ISAC workshop are focused on open and intelligent RAN, AI RAN development, and upcoming 6G standardization. For wireless researchers, this is the kind of process signal that matters: ideas around AI-native networking are being shaped inside the venues that eventually influence deployable architectures and interoperable ecosystems.

Source: https://www.o-ran.org/event/o-ran-alliance-f2f-meeting-1-5-june-2026-seattle

Nokia is framing the AI boom as fundamentally a networking and infrastructure challenge

Nokia’s first global AI Summit of 2026 brought together more than 200 leaders around the idea that the emerging AI supercycle is inseparable from connectivity. Underneath the branding is a useful industry read: telecom vendors increasingly want to define themselves as the physical and operational substrate beneath model deployment, data-center growth, and autonomous systems.

Source: https://x.com/nokia/status/2062082572268491022

Research Radar

Statistically Robust Resource Block Allocation for Satellite Communications

Authors: Chaitanya Manapragada, Laurent Decreusefond, Philippe Martins
Venue: arXiv
This paper develops a footprint-level rule for sizing resource blocks in satellite systems under uncertain, spatially correlated attenuation. Its value is practical: it gives operators both a simulation-driven estimate and a conservative analytical bound for overload probability before expensive deployments are locked in.

Source: https://arxiv.org/abs/2606.02124

GNN-based Online Beamforming Design for HAPS-Assisted NTN

Authors: Lavanya S S Anjapuli, Animesh Yadav, Halim Yanikomeroglu
Venue: arXiv
The authors propose a HAPS-assisted non-terrestrial architecture and use a graph-neural-network optimizer for online beamforming. The main implication is better energy efficiency and stronger support for cell-edge users in mixed terrestrial/non-terrestrial settings.

Source: https://arxiv.org/abs/2606.00244

6G Needs Agents: Toward Agentic AI-Native Networks for Autonomous Intelligence

Authors: Mohamed Amine Ferrag, Abderrahmane Lakas, Merouane Debbah
Venue: arXiv
This paper argues that 6G should not stop at optimization loops and instead should incorporate bounded, policy-governed LLM agents across device, edge, and core layers. The useful contribution is not hype but architectural framing: it identifies the tradeoff between reasoning quality, efficiency, and deployment placement across the network continuum.

Source: https://arxiv.org/abs/2605.01546

MIT/Harvard Events This Week

• Wed, June 3 — Intro to MIT’s AI Tools @ MIT, 600 Technology Square, Cambridge
  Source: https://calendar.mit.edu/event/intro-to-mits-ai-tools

• Wed, June 3 — Advanced Claude Code: Skills, MCPs, Hooks, and Multi-Agent Workflows @ Harvard Bok Center, 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/advanced-claude-code-skills-mcps-hooks-and-multi-agent-workflows-0?occ_id=0

• Thu, June 4 — Claude Code Capstone: Build, Vibecheck, Deploy @ Harvard Bok Center, 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/claude-code-capstone-build-vibecheck-deploy-0?occ_id=0

Source Issues

• TNT’s calendar page was still stale and mostly surfaced February–April listings, so direct MIT/Harvard event pages were used instead.
• arXiv’s API returned 429 rate-limit errors during discovery, so paper selection fell back to direct paper pages and web search.
• Next G Alliance and several rotated X accounts were stale, so the X pass leaned more on Nokia plus official Microsoft, GitHub, and Anthropic sources.

Takeaway

The cleanest pattern this morning is that AI competition is shifting from raw model launches toward secure, governed agent infrastructure across policy, developer tools, telecom, and scientific R&D.