Anthropic raises a massive Series H round

Anthropic announced on May 28 that it raised $65 billion at a $965 billion post-money valuation. The company said the money will fund safety and interpretability research, compute expansion, and broader product and partnership scaling for Claude.

What matters is not just the round size. Frontier labs are increasingly being capitalized like infrastructure companies, where access to compute, distribution, and operating capacity is becoming a strategic moat.

Source: https://www.anthropic.com/news/series-h

OpenAI publishes a Frontier Governance Framework

OpenAI released a public Frontier Governance Framework describing how its safety and security practices align with emerging requirements such as California’s Transparency in Frontier AI Act and the EU AI Act’s Code of Practice for General Purpose AI.

This is a sign that frontier model governance is becoming more formalized and externally legible. Labs are no longer relying only on internal preparedness memos; they are increasingly publishing structured governance documents that map to law and policy.

Source: https://openai.com/index/openai-frontier-governance-framework/

OpenAI details self-improving tax agents built with Codex

OpenAI published a case study on building self-improving tax agents with Codex. The workflow uses agents to extract values, compare predicted results against filed returns, and feed discrepancies back into the system for iterative improvement.

This is one of the clearest recent examples of agent systems moving beyond demos into measurable back-office automation with real correction loops and domain-specific feedback.

Source: https://openai.com/index/building-self-improving-tax-agents-with-codex/

Google pushes its newest image models into general availability

Google DeepMind said on X that Nano Banana 2 and Nano Banana Pro are now generally available to developers.

The bigger takeaway is platform normalization. Image generation is increasingly being treated as a standard capability inside model platforms rather than a standalone novelty feature.

Source: https://x.com/GoogleDeepMind/status/2060036701507010721

Qualcomm frames 6G as an AI-native platform

In a new May 28 article, Qualcomm argued that 6G should be designed from the start as an AI-native system with integrated compute, sensing, uplink gains, and context-aware control across device, RAN, and core.

For wireless researchers, this is a useful signal from industry: the 6G conversation is no longer just about a faster air interface. It is about building a distributed compute-network fabric that can support agentic services and real-time adaptation.

Source: https://www.qualcomm.com/news/onq/2026/05/6g-foundry-ai-native-platform

TSMC says AI power demand is reshaping chip design

Reuters reported that a senior TSMC executive said surging electricity demand from AI is making energy efficiency the dominant constraint in future chip design, overtaking raw compute growth as the primary pressure.

That matters because it reframes the semiconductor race. Future winners may be defined less by peak performance alone and more by how efficiently they convert power into useful AI work.

Source: https://www.marketscreener.com/news/energy-use-forcing-rethink-of-ai-chip-design-tsmc-says-ce7f5ddadf89f220

Research Radar

Intent-Based Orchestration in Open RAN: An ns-3 Simulation Framework

Authors: Pouya Agheli, Grégoire Lefebvre
Venue: EuCNC & 6G Summit 2026

This paper introduces an ns-3-based framework for evaluating intent-based, semantics-aware control in Open RAN. It shows that intent-based radio resource management can improve an Intent Satisfaction Score while lowering radio-resource usage and compute overhead.

Source: https://arxiv.org/abs/2605.30079

ARIADNE: AI-RAN Informed Link Adaptation in Digital Twin Network Environments

Authors: Maria Tsampazi, Neagin Neasamoni Santhi, Nicole Perrotta, Falko Dressler, Tommaso Melodia
Venue: arXiv

ARIADNE uses reinforcement learning in a digital-twin environment for link adaptation and reports improvements in spectral efficiency over industry-standard and recent baselines. It is a good example of AI-RAN methods being tested in realistic, pre-deployment environments.

Source: https://arxiv.org/abs/2605.29772

A Preliminary Assessment of Midhaul Links at 140 GHz using Ray-Tracing

Authors: Sravan Reddy Chintareddy, Marco Mezzavilla, Sundeep Rangan, Morteza Hashemi
Venue: arXiv

This paper studies 140 GHz wireless midhaul in urban 5G transport architectures and finds the band promising for high-layer split midhaul links between central and distributed units.

Source: https://arxiv.org/abs/2605.27771

MIT/Harvard Events This Week

• June 1, 2026 — Getting Started with Claude and Cowork @ Harvard Bok Center, 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/getting-started-claude-and-cowork-0?occ_id=0

• June 2, 2026 — Claude Code: Setup, Commands, and Context @ Harvard Bok Center, 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/claude-code-setup-commands-and-context-0?occ_id=0

• June 3, 2026 — Intro to MIT’s AI Tools @ 600 Technology Square, Cambridge
  Source: https://calendar.mit.edu/event/intro-to-mits-ai-tools

• June 3, 2026 — 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

Source Issues

• TNT’s calendar page was stale and mostly surfaced February–April entries, so direct MIT and Harvard event pages were used instead.
• arXiv API requests were rate-limited this morning, so paper discovery used the recent-list pages and direct abstract pages.
• X worked, but several rotated accounts were stale, promotional, or duplicates of topics already covered in the prior three digests.

Takeaway

Today’s strongest pattern is operationalization: frontier AI is being financed, governed, productized, and increasingly constrained by the physical realities of networks, chips, and power.

Morning Digest — Thursday, May 28, 2026

1) Anthropic says stronger agents need stronger containment

Anthropic’s engineering team argues that agent permissions and sandbox boundaries should scale with model capability rather than stay static. That is a meaningful shift in emphasis: frontier model safety is no longer just about evals and red-teaming, but about how much real-world authority an agent receives in deployment.

Source: https://www.anthropic.com/engineering/contain-claude

2) Project Glasswing has already surfaced 10,000+ severe software flaws

Anthropic’s latest Project Glasswing update says the initiative and its partners have already found more than ten thousand high- or critical-severity vulnerabilities in essential software. If the detection side of cyber keeps accelerating, the limiting factor may become patch throughput, triage, and operational response capacity.

Source: https://www.anthropic.com/research/project-glasswing-update

3) Google is expanding SynthID verification into Search and Chrome

Google DeepMind says SynthID has now watermarked more than 100 billion pieces of content and is widening verification into Search and Chrome while bringing in OpenAI, ElevenLabs, and Kakao as partners. The important signal is not just watermark scale, but that provenance checks are being woven into everyday consumer surfaces.

Source: https://x.com/GoogleDeepMind/status/2059235181274202500

4) GitHub rotated signing keys after a Spotlight package-publishing incident

GitHub published an important security update saying it revoked affected keys and rotated credentials after a compromise tied to the Spotlight package-publishing pipeline. It is a reminder that software supply-chain resilience still depends heavily on fast incident containment and trust repair.

Source: https://github.blog/changelog/2026-05-26-important-security-update-for-the-spotlight-package/

5) GitHub Copilot now supports remote coding sessions

GitHub announced remote coding sessions for Copilot, letting users monitor and redirect coding agents away from the machine where they started. This matters because coding agents are gradually turning from local IDE helpers into persistent, supervisable work surfaces.

Source: https://github.blog/changelog/2026-05-27-control-your-coding-agent-from-anywhere-with-remote-coding-sessions-in-github-copilot/

6) Starlink is validating next-generation V3 hardware in orbit

Starlink says modified satellites carrying V3 components were deployed during Starship’s latest flight test, giving SpaceX an on-orbit path to test next-generation constellation hardware before broader rollout. For LEO infrastructure, this is one of the clearest signs that SpaceX is iterating not just on launch cadence but on constellation architecture itself.

Source: https://x.com/Starlink/status/2057959999629033547

Research Radar

Low-Altitude Wireless Networks: The Next Horizon of Wireless Infrastructure — Jiali Nie, arXiv

This paper frames low-altitude airspace as a new three-dimensional wireless domain and introduces LAWN as a tightly coupled communication-sensing-control architecture. It is interesting for 6G because it treats airspace capacity and wireless capacity as a single co-designed systems problem.

🔗 https://arxiv.org/abs/2605.24368

Secure UAV Swarms in Low-Altitude Wireless Networks: Challenges and Solutions — Yuntao Wang, arXiv

This work proposes a cloud-edge-end collaborative defense framework for UAV swarms, spanning GPS spoofing resistance, trust-based authentication, and multi-agent attack forensics. It is a useful snapshot of how aerial networking research is merging security and autonomy rather than treating them as separate layers.

🔗 https://arxiv.org/abs/2605.26876

MIT/Harvard Events This Week

• Thu, May 28 — Gen AI and Its Impact on the World: Step into the Future of Innovation @ Harvard University
  Source: https://careerservices.fas.harvard.edu/events/2026/05/28/gen-ai-and-its-impact-on-the-world-step-into-the-future-of-innovation/

• Mon, June 1 — Getting Started with Claude and Cowork @ Harvard Bok Center, 50 Church Street
  Source: https://bokcenter.harvard.edu/event/getting-started-claude-and-cowork-0?occ_id=0

• Wed, June 3 — Intro to MIT’s AI Tools @ MIT Building NE49, 405
  Source: https://calendar.mit.edu/event/intro-to-mits-ai-tools

Source Issues

• TNT’s calendar page still appears stale and mostly shows February–April listings, so event picks were cross-checked on direct Harvard and MIT pages.
• arXiv API requests timed out this morning, so research selection fell back to direct arXiv pages and search results.
• X access worked, but AST SpaceMobile returned no tweets in today’s rotated pass.

Takeaway

Containment, provenance, supply-chain trust, remote agent supervision, and orbit-side hardware validation are becoming core infrastructure layers rather than side topics.

Morning Digest — Wednesday, May 27

Telesat is chasing a secure satellite deal with Italy

Reuters reports that Canada’s Telesat held preliminary talks with Italy about encrypted satellite communications for government, diplomatic, and defense use. This is worth watching because it pushes LEO competition beyond consumer broadband and into sovereign communications, where procurement cycles are slower but contracts can be strategically sticky.

Source: https://www.reuters.com/business/media-telecom/canadas-telesat-eyes-secure-italian-satellite-connectivity-contract-sources-say-2026-05-26/

Europe’s next mobile-satellite spectrum plan may leave room for Starlink and Amazon

A Reuters exclusive says European companies are set to receive most of the future mobile-satellite spectrum allocation, with the rest potentially open to outside players like Starlink and Amazon. The big implication is that non-terrestrial networking is becoming a spectrum-governance story as much as a launch and hardware story.

Source: https://www.reuters.com/business/aerospace-defense/european-companies-set-receive-two-thirds-future-mobile-satellite-spectrum-rest-2026-05-26/

SK hynix joined the $1 trillion club after Samsung and Micron

Reuters says SK hynix crossed the $1 trillion market-cap mark, following Samsung and Micron as investors keep rewarding AI-memory winners. That reinforces a useful market truth: AI infrastructure is not only about model labs and GPU vendors anymore; memory suppliers are now central to the bottleneck stack.

Source: https://www.reuters.com/world/asia-pacific/sk-hynix-market-capitalisation-tops-1-trln-2026-05-27/

American Airlines is bringing Starlink onboard its flights

Starlink said it will provide in-flight internet for American Airlines, extending LEO connectivity deeper into mainstream passenger infrastructure. Airline Wi-Fi has long been a pain point, so this is one of those mundane-but-important deployments that could shift user expectations if performance holds up.

Source: https://x.com/Starlink/status/2059292635911819641

Starlink-supported buoys streamed live video during Starship’s latest flight test

Starlink said ocean buoys in the Indian Ocean delivered real-time video streaming during Starship’s twelfth flight test. The interesting part is not just the space spectacle, but the operational proof that LEO links can support hard-to-wire telemetry setups in remote marine environments.

Source: https://x.com/Starlink/status/2059391182094880785

NVIDIA is pushing fresh Vera CPU benchmark claims for agentic AI workloads

NVIDIA highlighted new Phoronix benchmark results for its Vera CPU, claiming major gains in overall performance, Linux kernel compilation, and memory bandwidth for AI-factory workloads. Even though Vera is not brand-new news, the updated benchmark push shows how NVIDIA is framing agentic AI as a full-system compute problem rather than a pure GPU story.

Source: https://x.com/nvidia/status/2059387019109826953

Google DeepMind spotlighted Gemini Embedding 2 as its multimodal retrieval layer

Google DeepMind is continuing to push Gemini Embedding 2, its multimodal embedding model for retrieval, classification, clustering, and recommendations across text, images, video, audio, and documents. This matters because better embeddings quietly improve the grounding layer that powers search, RAG, and agent memory in production systems.

Source: https://deepmind.google/models/gemini/embedding/

Research Radar

GENESIS: Harnessing AI Agents for Autonomous 6G RAN Synthesis, Research, and Testing

Authors: Tamerlan Aghayev, Maxime Elkael, Michele Polese, Minh Dat Nguyen, Gabriele Gemmi, Andrea Lacava, Ali Saeizadeh, Reshma Prasad, Paolo Testolina, Angelo Feraudo, Soumendra Nanda, Pedram Johari, Salvatore D’Oro, Tommaso Melodia
Venue: arXiv

GENESIS proposes an agentic framework for cellular R&D that converts intents such as standards clauses, telemetry anomalies, and research hypotheses into over-the-air validated solutions. For Dad’s lane, the key idea is compounding RAN experimentation through reusable skills and a persistent knowledge layer instead of one-off manual engineering loops.

Source: https://arxiv.org/abs/2605.27360

On the LEO Satellite Constellation Design for North Atlantic Coverage

Authors: Alejandro Ramírez-Arroyo, Miguel Villanueva-Fernández, Preben Mogensen
Venue: arXiv

This paper studies how inclination, minimum elevation angle, altitude, and footprint affect visibility probability, revisit time, path loss, and coverage continuity for North Atlantic service. It is especially relevant for maritime, aviation, and Arctic scenarios where generic global-coverage assumptions are too coarse.

Source: https://arxiv.org/abs/2605.26943

OpenTwin: Digital Twin Driven Closed Loop KPM Inference and Control for Open RAN

Authors: Md Sharif Hossen, Zifan Zhang, Dara Ron, Yuchen Liu, Vijay K. Shah
Venue: arXiv

OpenTwin builds an O-RAN digital twin that learns live network behavior, corrects deviations against real measurements, and validates control policies before applying them to physical infrastructure. Reported fidelity up to 96% makes this a practical digital-twin paper rather than a vague architectural sketch.

Source: https://arxiv.org/abs/2605.24662

MIT/Harvard Events This Week

• Thu, May 28 — Gen AI and Its Impact on the World: Step into the Future of Innovation @ Harvard University
  Source: https://careerservices.fas.harvard.edu/events/2026/05/28/gen-ai-and-its-impact-on-the-world-step-into-the-future-of-innovation/

• Mon, June 1 — Claude Code: Putting it all together to develop course content and communicate research @ Harvard Bok Center, 50 Church Street
  Source: https://bokcenter.harvard.edu/generative-ai-events

• Wed, June 3 — Intro to MIT’s AI Tools @ MIT Building NE49, 405
  Source: https://calendar.mit.edu/event/intro-to-mits-ai-tools

Source Issues

• TNT’s calendar page still appears stale and mostly shows February–April listings, so event picks were cross-checked on direct Harvard/MIT pages.
• arXiv API requests returned 429 errors this morning, so paper discovery fell back to arXiv recent-list pages and abstract pages.
• OpenAI and AST SpaceMobile did not yield fresh usable items in the rotated X pass, and Next G Alliance’s feed appeared stale.

Takeaway

The frontier keeps looking more infrastructural: satellite spectrum, sovereign communications, AI memory hardware, multimodal retrieval, and digital-twin control loops are all becoming more strategically important than one more flashy model demo.

Morning Digest — Tuesday, May 26

Google DeepMind’s AlphaProof Nexus pushes AI math into proof-checked research

Google DeepMind’s new paper on AlphaProof Nexus reports that the system autonomously solved 9 open Erdős problems and proved 44 OEIS conjectures by combining language-model generation with Lean-based formal verification. The bigger story is methodological: this is a stronger template for using AI in research settings where correctness matters, because the proof assistant acts as a hard constraint rather than a soft reviewer.

Source: https://arxiv.org/abs/2605.22763

OpenAI is making Codex more mobile and context-aware

OpenAI’s latest Codex updates add Appshots on macOS and mobile preview support, letting users attach app context to a thread and stay connected to ongoing work from a phone while a Mac host keeps running. That nudges coding agents toward a persistent workflow model instead of a single-terminal interaction.

Source: https://help.openai.com/en/articles/6825453-chatgpt-release-notes

Anthropic buys Stainless to strengthen agent connectivity

Anthropic acquired Stainless, the SDK and MCP tooling company behind its official SDKs. This is an infrastructure move with outsized implications: as models improve, the differentiator increasingly becomes which systems agents can safely and reliably access.

Source: https://www.anthropic.com/news/anthropic-acquires-stainless

Anthropic and KPMG are scaling Claude across a global professional-services giant

Anthropic says KPMG is deploying Claude across its core business and workforce of more than 276,000 through a strategic alliance focused on governed enterprise adoption. It’s a useful signal that large advisory firms are trying to normalize AI not as a pilot, but as a standard layer in delivery and internal operations.

Source: https://www.anthropic.com/news/anthropic-kpmg

Google’s Gemini eyewear push turns AI assistants into something you actually wear

Google says intelligent eyewear built with Samsung and frame partners Gentle Monster and Warby Parker is arriving this fall. Gemini will handle directions, messages, photos, and in-the-moment assistance, making this one of the more serious mainstream attempts to turn AI into ambient, wearable computing.

Source: https://blog.google/products-and-platforms/platforms/android/android-xr-io-2026/

Starlink expands into Kyrgyzstan

Starlink announced availability in the Kyrgyz Republic. These country-by-country activations rarely dominate headlines, but they matter because they show the slow compounding of LEO internet into real global access infrastructure.

Source: https://x.com/Starlink/status/2058977657015140841

Qualcomm keeps pushing agentic RAN as the bridge to AI-native 6G

Qualcomm’s Agentic RAN Management Service and associated commercial RAN AI features remain worth attention because they frame present-day automation as groundwork for the 6G era. For wireless researchers, it’s a concrete example of vendors trying to turn the “AI-native network” idea into actual operational software.

Source: https://www.qualcomm.com/news/releases/2026/03/qualcomm-launches-agentic-ran-management-service-and-ai-enhancem

Research Radar

Towards Resilient and Autonomous Networks: A BlueSky Vision on AI-Native 6G

Liang Wu, Kelly Wan, Mayank Darbari, Liangjie Hong

This paper sketches a 6G architecture centered on a foundation model plus collaborative multi-agent systems, treating network management as a unified multimodal optimization problem rather than a pile of isolated ML tasks.

Source: https://arxiv.org/abs/2605.21395

Toward the Internet of Space Things: Performance Analysis of LEO Satellite Relay Networks using mmWave and sub-THz links

Sergi Aliaga, Ahmad Masihi, Vitaly Petrov, Marc Sanchez Net, Josep M. Jornet

A strong LEO paper for Dad’s interests: it analyzes inter-satellite relay backbones for “space users” and argues that mmWave/sub-THz links could unlock major gains in contact probability, channel capacity, and continuous connectivity.

Source: https://arxiv.org/abs/2605.02061

Toward LEO Satellite Network Systems for Instantaneous Detection of Environmental Changes

Zian Wang, Peng Hu, Grant Gunn

This work explores orbital edge computing for real-time environmental monitoring and shows that some constellation designs can keep age-of-information low enough for sub-minute situational awareness.

Source: https://arxiv.org/abs/2605.01243

MIT / Harvard Events This Week

• Tue, May 26 — 30 Years Since the Telecommunications Act: A Summit on Promise, Progress, and the Work Ahead @ MIT Media Lab, E14 Atrium
  Source: https://calendar.mit.edu/event/30-years-since-the-telecommunications-act-a-summit-on-promise-progress-and-the-work-ahead

• Tue, May 26 — Art/Science lunch with Felice Frankel @ MIT Building 68-121
  Source: https://calendar.mit.edu/event/artscience-lunch-with-felice-frankel-526-1230p-in-68-121-hosted-by-the-mit-biology-artists-resource-collective

• Mon, June 1 — Getting Started with Claude and Cowork @ Harvard Bok Center, 50 Church Street
  Source: https://bokcenter.harvard.edu/generative-ai-events

• Wed, June 3 — Intro to MIT’s AI Tools @ MIT Building NE49, 405
  Source: https://calendar.mit.edu/event/intro-to-mits-ai-tools

Source Issues

• TNT’s calendar page still appears stale and mostly shows February–April entries, so MIT and Harvard event picks were cross-checked on direct event pages.
• arXiv API requests returned 429 errors this morning, so paper discovery fell back to search-indexed arXiv pages.
• X access worked, but several rotated accounts only had promotional or already-covered items.
• /Users/bruski/clawspace/memory/2026-05-26.md did not exist yet, so short-term recall used the May 25 memory file plus archive-based deduping.

Takeaway

The pattern this morning is infrastructure hardening: AI is getting more trustworthy in research, more ambient in devices, more embedded in enterprise systems, and more tightly coupled to the networks beneath it.

Morning Digest — Monday, May 25

OpenAI says a general reasoning model cracked a classic geometry conjecture

OpenAI published a new research note arguing that one of its general-purpose reasoning models disproved a long-standing conjecture in the planar unit distance problem, a famous question in discrete geometry first posed by Paul Erdős. The company says external mathematicians checked the proof and that the result provides one of the clearest cases yet of a frontier model contributing directly to original mathematical research rather than simply assisting with exposition or search.

Why it matters: if this result stands, it strengthens the case that advanced reasoning models can become genuine research collaborators in narrow but meaningful domains where correctness can be independently verified.

Source: https://openai.com/index/model-disproves-discrete-geometry-conjecture/

Google opens Gemini for Science tools to researchers

Google introduced Gemini for Science, a new collection of experimental tools intended to support scientific work with hypothesis generation, computational discovery, and literature synthesis. The company is gradually opening access through Google Labs, and the framing is notable: Google is explicitly pushing multi-agent scientific workflows, not just generic chat interfaces, as the next interface layer for research.

Why it matters: this is part of a broader shift from “AI as assistant” toward “AI as structured research engine,” especially for large-scale literature synthesis and parallelizable computational exploration.

Source: https://blog.google/innovation-and-ai/technology/research/gemini-for-science-io-2026/

IBM and the U.S. Commerce Department back a dedicated American quantum foundry

IBM and the U.S. Department of Commerce announced plans for Anderon, a standalone 300mm quantum-wafer foundry supported by a proposed $1 billion CHIPS incentive and another $1 billion from IBM. According to IBM, the goal is to create America’s first pure-play quantum foundry and build domestic manufacturing capacity for multiple quantum hardware vendors.

Why it matters: this is a meaningful move from pure quantum R&D signaling into industrial infrastructure, supply chain strategy, and national manufacturing policy.

Source: https://newsroom.ibm.com/ibm-and-u-s-department-of-commerce-announce-americas-first-purpose-built-quantum-foundry

Intel is pitching CPU+GPU+NPU robots as a cheaper edge-AI stack

Intel says Sensory AI’s Ella robotic system now runs on Intel Core Ultra Series 3 without relying on a discrete GPU, using the chip’s integrated CPU, GPU, and NPU mix to support multiple agents at the edge. Intel’s pitch is that physical AI deployments need lower thermals, lower cost, and easier field maintenance more than they need oversized cloud-style acceleration.

Why it matters: for real-world edge robotics, the most important battle may be total cost of ownership and operational simplicity, not peak benchmark numbers.

Source: https://newsroom.intel.com/artificial-intelligence/intel-core-ultra-series-3-for-edge-ai-robotics

Nokia launches a physics-based RAN digital twin for AI-native 6G

Nokia launched a RAN Digital Twin powered by NVIDIA Aerial Omniverse Digital Twin, targeting realistic simulation of radio propagation, beamforming, mobility, and device-specific behavior in dense environments. Nokia positions the system as a core tool for designing and optimizing AI-native 6G networks before deployment.

Why it matters: for 6G and advanced wireless research, this is exactly the kind of train-simulate-deploy infrastructure that could shrink concept-to-live cycles and reduce the cost of testing high-complexity radio scenarios.

Source: https://www.nokia.com/blog/nokia-launches-nokia-ran-digital-twin-to-turbo-charge-ai-native-6g-powered-by-nvidia-aerial-omniverse-digital-twin/

Microsoft and EY are scaling from AI pilots to enterprise operations

Microsoft says EY is expanding Microsoft 365 Copilot to more than 400,000 employees and that the two companies are jointly investing over $1 billion to move organizations from isolated AI pilots into repeatable enterprise transformation. The post highlights concrete workflow gains in finance, assurance, and tax operations, with Microsoft increasingly selling governance, rollout discipline, and operational integration as the differentiator.

Why it matters: the AI market is maturing from experimentation toward measurable business deployment, and the winners may be those who can operationalize multi-agent systems inside large organizations.

Source: https://blogs.microsoft.com/blog/2026/05/21/from-ai-pilots-to-enterprise-impact-why-execution-is-the-new-differentiator/

Research Radar

Physics-Informed Digital Twins for Channel Estimation and Traffic Prediction of Non-Terrestrial Networks

Authors: Xinyu Huang, Yixiao Zhang, Xue Qin, Mingcheng He, Junling Li, Weihua Zhuang, Xuemin Shen
Venue: arXiv (eess.SP)
This paper proposes a physics-informed digital-twin framework for non-terrestrial networks that reconstructs full-resolution CSI from sparse outdated pilots and predicts traffic using an orbit-adaptive spatiotemporal graph model. The evaluation uses a high-fidelity simulation environment built on real Starlink ephemeris, global population, and weather data.

🔗 https://arxiv.org/abs/2605.23155

6G Communication Networks Enabling Embodied Agents: Architecture and Prototype

Authors: Lipeng Dai, Luping Xiang, Kun Yang
Venue: arXiv (cs.RO / eess.SP)
This work proposes a hierarchical communication architecture for embodied agents, linking human-intent perception, O-RAN transport, an intelligent intermediary layer, and physical embodiment. The paper also reports an end-to-end prototype with millisecond-level latency and stable closed-loop operation.

🔗 https://arxiv.org/abs/2605.23263

AdaPTwin: Adaptive Multi-Fidelity Predictive Digital Twin for Proactive Radio Resource Management in Vehicular Networks

Authors: Armin Makvandi, Md. Zoheb Hassan, Md. Jahangir Hossain
Venue: arXiv (eess.SY / cs.NI)
AdaPTwin introduces an adaptive multi-fidelity predictive network digital twin for vehicular radio resource management, dynamically changing simulation fidelity based on network conditions. The reported gains over non-adaptive baselines are substantial, including major improvements in sum-rate and outage probability.

🔗 https://arxiv.org/abs/2605.21897

MIT / Harvard Events This Week

• Tue, May 26 — 30 Years Since the Telecommunications Act: A Summit on Promise, Progress, and the Work Ahead @ MIT Media Lab, E14 Atrium
  Source: https://calendar.mit.edu/event/30-years-since-the-telecommunications-act-a-summit-on-promise-progress-and-the-work-ahead

• Tue, May 26 — Art/Science lunch with Felice Frankel @ MIT Building 68-121
  Source: https://calendar.mit.edu/event/artscience-lunch-with-felice-frankel-526-1230p-in-68-121-hosted-by-the-mit-biology-artists-resource-collective

• Mon, June 1 — Getting Started with Claude and Cowork @ Harvard Bok Center, 50 Church Street
  Source: https://bokcenter.harvard.edu/event/getting-started-claude-and-cowork-0?occ_id=0

Source Issues

• TNT’s calendar page fetched successfully but appears stale, showing February/March 2026 entries rather than this or next week.
• arXiv API requests hit rate limits or timed out, so paper selection fell back to direct arXiv recent-category pages and abs pages.
• X access worked, but the rotated accounts did not surface stronger fresh leads than the primary-source articles above.

Takeaway

The strongest signal this morning is that AI is moving deeper into infrastructure: scientific discovery systems, digital twins for wireless networks, embodied edge compute, quantum manufacturing, and enterprise-scale execution are all getting more concrete at the same time.

Overview

Today’s pattern is grounding: frontier AI is getting tied to real software, real provenance systems, real environments, real schools, and more operational network stacks.

Top Stories

Anthropic says AI-assisted vulnerability hunting is already outrunning patch capacity

Anthropic’s initial Project Glasswing update says Claude Mythos Preview and roughly 50 partners have already surfaced more than 10,000 high- or critical-severity vulnerabilities in important software. The interesting shift is that cyberdefense may no longer be bottlenecked on finding vulnerabilities first, but on verifying, disclosing, and patching them fast enough.

Source: https://www.anthropic.com/research/glasswing-initial-update

Google expands SynthID and content-verification tools across its products

Google says it is broadening its provenance stack across Search, Gemini, Chrome, Pixel, and Cloud, while expanding partner use of SynthID watermarking. That matters because AI-origin detection is moving from a specialized trust-and-safety feature toward something that looks like mainstream web infrastructure.

Source: https://blog.google/innovation-and-ai/products/identifying-ai-generated-media-online/

Google connects Project Genie to Street View

Google DeepMind says Project Genie can now turn real U.S. places from Street View into interactive simulated worlds for eligible AI Ultra users. This is notable because world models are becoming less toy-like and more anchored to real geospatial context, which could matter for simulation, training, and embodied-agent workflows.

Source: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/project-genie-expands/

Starlink says 140 schools in Brazil’s Amazon region are now online

Starlink says its collaboration with the Brazilian non-profit Redes do Futuro has connected 140 schools across the remote Amazon region, serving more than 14,000 students. For the LEO narrative, this is one of the clearest examples of satellite connectivity showing up as basic civic infrastructure rather than just a launch or direct-to-cell headline.

Source: https://x.com/Starlink/status/2058267430829670801

Qualcomm spotlights a new ONNX Runtime plugin path for edge AI deployment

Qualcomm’s weekly AI roundup says it debuted the first ONNX Runtime Plugin Execution Provider and highlighted new AI Hub deployments including CenterPoint for real-time LiDAR object detection at the edge. The practical signal is that model deployment across embedded and automotive systems is getting more productized and less bespoke.

Source: https://x.com/Qualcomm/status/2057952187850895810

Andrej Karpathy joins Anthropic

Karpathy said he has joined Anthropic and expects the next few years at the LLM frontier to be especially formative. Talent moves at this level are worth tracking because they often precede shifts in research culture, technical direction, and the educational layer around frontier-model development.

Source: https://x.com/karpathy/status/2056753169888334312

DeepSeek makes its V4-Pro discount permanent

DeepSeek said the discounted pricing for DeepSeek-V4-Pro is now permanent rather than a temporary promotion. That is a meaningful market signal for builders, because the next phase of model competition increasingly depends on durable cost-performance and long-context practicality rather than only leaderboard prestige.

Source: https://x.com/deepseek_ai/status/2057854261699195173

Research Radar

Beyond Spherical Wavefront: Near-Field Channel Estimation Under Wavefront Anisotropy

Authors: Heling Zhang, Xiujun Zhang, Xiaofeng Zhong, Shidong Zhou
Venue: arXiv
This paper targets a very practical 6G/mmWave problem: near-field channel estimation breaks down when reflected wavefronts stop behaving like nice clean spheres. The authors propose an anisotropic wavefront model and corresponding estimation method, which could matter for precise beamforming as arrays get larger and environments get messier.

Source: http://arxiv.org/abs/2605.22117v1

A Camera-Cooperative ISAC Framework for Multimodal Non-Cooperative UAVs Sensing

Authors: Wenfeng Wu, Luping Xiang, Kun Yang
Venue: arXiv
This one is interesting because it blends cameras with integrated sensing-and-communications (ISAC): vision handles coarse airspace awareness while wireless sensing handles precise tracking. That multimodal division of labor feels realistic for UAV monitoring systems that need both efficiency and accuracy.

Source: http://arxiv.org/abs/2605.22090v1

Demystifying Deep Reinforcement Learning: A Neuro-Symbolic Framework for Interpretable Open RAN Automation

Authors: Jie Lu, Peihao Yan, Pang-Ning Tan, Y. Thomas Hou, Huacheng Zeng
Venue: arXiv
Open RAN automation keeps running into the trust problem: DRL can work, but operators do not want opaque control logic inside carrier-grade systems. This paper proposes a neuro-symbolic distillation approach to turn black-box DRL policies into more interpretable symbolic rules, which is exactly the kind of bridge real deployments need.

Source: http://arxiv.org/abs/2605.10648v2

MIT/Harvard Events This Week

• June 1 — Getting Started with Claude and Cowork @ 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/getting-started-claude-and-cowork-0?occ_id=0
• June 2 — Claude Code: Setup, Commands, and Context @ 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/claude-code-setup-commands-and-context-0?occ_id=0
• June 3 — Intro to MIT’s AI Tools @ Building NE49, 405, 600 Technology Square, Cambridge
  Source: https://calendar.mit.edu/event/intro-to-mits-ai-tools

Source Issues

• TNT calendar still returned stale February–April listings, so event picks were cross-checked on direct Harvard/MIT pages.
• @ASTSpaceMobile returned no fresh usable posts, and @NextGAlliance was effectively stale.
• IEEE Xplore and ACM did not surface strong fresh matches quickly, so today’s Research Radar leans on arXiv.

Bottom Line

Frontier AI is becoming more operational and more grounded: in security workflows, provenance systems, real-world simulation, edge deployment stacks, and connectivity infrastructure that reaches actual communities.

Overview

Today’s pattern is convergence: AI agents, next-gen wireless, vehicles, maritime systems, and orbital infrastructure are all being packaged as deployable platforms rather than isolated demos.

Top Stories

1) Starship’s twelfth test delivered a real orbital-infrastructure milestone

SpaceX said Starship V3 flew for the first time, deployed modified Starlink satellites, and executed a return-profile banking maneuver before splashdown. This makes the flight relevant not just as launch spectacle, but as a live test of heavy-lift deployment mechanics that could matter for future large-scale orbital networks and Starlink expansion.

Source: https://x.com/SpaceX/status/2058005949466501590

2) OpenAI gets a fresh enterprise signal in coding agents

OpenAI said Gartner placed it as a Leader in the 2026 Magic Quadrant for Enterprise AI Coding Agents, with Codex now used by more than 4 million people weekly. The larger signal is that coding agents are being repositioned from productivity assistants into governed enterprise infrastructure with auditability, sandboxing, and deployment controls.

Source: https://openai.com/index/gartner-2026-agentic-coding-leader/

3) Google expands its AI partnership with Singapore

Google said it is broadening work with the Singapore government on scientific discovery, pandemic preparedness, and healthcare, supported by Google DeepMind’s local presence. This matters because national AI partnerships are evolving from vague training initiatives into sector-specific public-interest deployment programs.

Source: https://blog.google/company-news/inside-google/around-the-globe/google-asia/singapore-government-partnership/

4) Qualcomm and Stellantis deepen the vehicle-AI stack

Qualcomm and Stellantis expanded their collaboration to standardize Snapdragon Digital Chassis platforms across next-generation vehicle architectures, including cockpit, connectivity, and ADAS systems. The mention that Stellantis-owned aiMotive may join Qualcomm Technologies, subject to conditions, suggests growing vertical consolidation around shared compute and autonomy platforms.

Source: https://www.qualcomm.com/news/releases/2026/05/stellantis-and-qualcomm-expand-partnership-to-adopt----snapdrago

5) The U.S. and Sweden sign a tech deal centered on AI, 5G/6G, and semiconductors

The White House published a new Technology Prosperity Deal with Sweden that explicitly calls for stronger joint R&D on 5G/6G, wireless networks, cloud, electronics, security, and advanced materials relevant to semiconductor supply chains. For wireless and industrial-policy watchers, it is a clear sign that connectivity and AI are being framed as strategic alliance infrastructure.

Source: https://www.whitehouse.gov/releases/2026/05/technology-prosperity-deal-between-the-united-states-and-sweden/

6) Ericsson pushes 4G/5G plus agentic AI into maritime operations

Ericsson and Net Feasa announced a partnership to bring onboard 4G/5G connectivity and an agentic-AI-ready data layer to container fleets, with deployments already under way globally. The operational angle matters: this is less about abstract “smart shipping” and more about real-time reefer monitoring, dangerous-goods handling, early heat detection, and ship-to-shore data continuity.

Source: https://www.ericsson.com/en/press-releases/2026/5/ericsson-and-net-feasa-to-deliver-maritime-connectivity-competitive-advantage-with-4g-5g-and-agentic-ai

7) NVIDIA is turning COMPUTEX week into an AI-factory stage

NVIDIA opened live coverage for GTC Taipei at COMPUTEX, setting up Jensen Huang’s June 1 keynote around AI factories, scaling infrastructure, agentic AI, and physical AI. Even before product announcements land, the framing shows where NVIDIA wants the next compute cycle to be interpreted: not just faster chips, but full-stack industrial AI systems.

Source: https://blogs.nvidia.com/blog/nvidia-gtc-taipei-computex-2026-news/

Research Radar

Fluid RIS (FRIS)-Assisted Index Modulation for 6G Wireless Communications

Authors: Xusheng Zhu, Kai-Kit Wong, Sai Xu, Hao Xu, Wen Chen, Hyundong Shin
Venue: arXiv
This paper studies how fluid reconfigurable intelligent surfaces can support index modulation more reliably by optimizing codebooks for response-domain separability rather than raw layout diversity. It is a useful systems-oriented contribution for programmable wireless environments in 6G.

Source: https://arxiv.org/abs/2605.22508

SCALE: Sensitivity-Aware Federated Unlearning with Information Freshness Optimization for Mobile Edge Computing

Authors: Zihao Ding, Beining Wu, Jun Huang
Venue: arXiv (accepted by ICDCS)
This is worth watching because it blends federated unlearning with freshness-aware MEC optimization, moving the discussion closer to practical privacy-preserving edge AI systems.

Source: https://arxiv.org/abs/2605.22589

Joint Communication and Computation Scheduling for MEC-enabled AIGC Services: A Game-Theoretic Stochastic Learning Approach

Authors: Huaizhe Liu, Xinyi Zhuang, Jiaqi Wu, Yuan Luo, Bin Cao, Lin Gao
Venue: IEEE Internet of Things Journal technical report / arXiv
Interesting because it treats edge AIGC as a joint wireless-and-compute scheduling problem, which is exactly the kind of cross-layer framing that will matter as generative workloads move into mobile systems.

Source: https://arxiv.org/abs/2605.22277

MIT / Harvard Events This Week

• June 1 — Getting Started with Claude and Cowork @ 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/getting-started-claude-and-cowork-0?occ_id=0

• June 2 — Claude Code: Setup, Commands, and Context @ 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/claude-code-setup-commands-and-context-0?occ_id=0

• June 3 — Intro to MIT’s AI Tools @ Building NE49, 405, 600 Technology Square, Cambridge
  Source: https://calendar.mit.edu/event/intro-to-mits-ai-tools

Source Issues

• TNT calendar still returned stale February–April listings, so event picks were cross-checked on direct Harvard/MIT pages.
• IEEE Xplore and ACM did not surface strong fresh matches quickly, so today’s Research Radar leans on recent arXiv items and one IEEE IoT Journal technical-report posting.
• X access worked, but several checked accounts surfaced promotional or already-covered items, so they were excluded under the dedup rule.

Final Takeaway

The strongest through-line today is operationalization: the frontier is shifting from impressive components to integrated systems that can ship, govern, and scale.

☀️ Morning Digest — Friday, May 22

1) Telstra and Ericsson deepen 6G work

Telstra and Ericsson signed a letter of intent to collaborate on 6G research, trials, and 3GPP evolution. The agreement includes access to Ericsson’s 6G testbed in Sweden and reciprocal testing in Australian conditions through Telstra’s Innovation Centre on the Gold Coast.

Why it matters: this is the kind of operator-vendor work that turns “AI-native 6G” from slideware into something testable, especially around standards evolution, geography-aware validation, and eventual network-sensing use cases.

Source: https://www.ericsson.com/en/press-releases/7/2026/telstra-and-ericsson-sign-agreement-to-advance-6g-research-and-innovation

2) Starlink Mobile expands toward satellite-to-phone service in Panama

Starlink said Starlink Mobile is coming to Panama with +Móvil and framed the service around apps, video, voice, and messaging where terrestrial coverage is weak.

Why it matters: recent direct-to-cell messaging has often centered on texting, but this announcement broadens the service language toward fuller mobile connectivity. That makes it a useful datapoint for anyone tracking the commercialization path of satellite-cellular convergence.

Source: https://x.com/Starlink/status/2057539003281494418

3) Gemini 3.5 Flash launches as Google’s new agent-first workhorse

Google introduced Gemini 3.5 and kicked off the family with 3.5 Flash, describing it as its strongest agentic and coding model yet. Google says the model is optimized for complex long-horizon tasks and delivers 4x the output token speed of other frontier models.

Why it matters: the frontier race is no longer just about raw intelligence. Labs are now competing on whether a model can reliably sustain long-running tool use while remaining fast and cheap enough to actually deploy.

Source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/

4) Google opens its managed-agent infrastructure to developers

Google also launched Managed Agents in the Gemini API. With a single API call, developers can spin up an agent that reasons, uses tools, browses the web, executes code, and keeps state inside an isolated Linux environment.

Why it matters: this is a platform move. Instead of forcing every developer to build orchestration, sandboxes, and recovery loops from scratch, Google is productizing the underlying agent harness itself.

Source: https://blog.google/innovation-and-ai/technology/developers-tools/managed-agents-gemini-api/

5) OpenAI lets Codex keep working after the Mac locks

OpenAI’s May 21 release notes added “locked computer use” for eligible Mac Computer Use users, plus new Appshots, goal mode availability, and better browser annotations.

Why it matters: coding agents are being shaped into persistent supervised workers, not just chat-driven copilots. The locked-use detail is especially telling because it removes a practical friction point in longer autonomous coding or investigation sessions.

Source: https://help.openai.com/en/articles/6825453-chatgpt-release-notes

6) NVIDIA and Dell push the AI factory deeper into the agent era

At Dell Technologies World, NVIDIA and Dell pitched a refreshed AI Factory stack built around Vera Rubin systems, faster agent sandboxes, and lower cost-per-token inference. NVIDIA’s post emphasized that agentic AI and secure on-prem deployment are now central enterprise design targets.

Why it matters: for companies building internal agents, the infrastructure question is shifting from “which GPU?” to “what full stack supports secure model execution, fast data access, and durable multi-step workflows?”

Source: https://blogs.nvidia.com/blog/dell-technologies-agent-enterprise-ai/

7) Qwen3.7-Max tightens the open competitive field for agentic AI

Alibaba’s Qwen team highlighted that Qwen3.7-Max gained 4.8 points over Qwen3.6-Max-Preview on the Artificial Analysis Intelligence Index, with improvements in scientific reasoning, coding, and agentic tasks.

Why it matters: even if OpenAI, Anthropic, and Google remain ahead at the very top, the gap is narrowing enough that the China-side frontier deserves daily attention—especially for coding, multilingual use, and cost-sensitive deployment.

Source: https://x.com/Alibaba_Qwen/status/2057451488587423830

📡 Research Radar

EnCoR: An end-to-end architecture for simplifying cellular networks — Wesley Woo et al., arXiv

A fresh systems paper that argues for a cleaner end-to-end cellular architecture, which could matter if future mobile networks want to reduce legacy control-path complexity while becoming more programmable.

🔗 https://arxiv.org/abs/2605.22524

SCALE: Sensitivity-Aware Federated Unlearning with Information Freshness Optimization for Mobile Edge Computing — Zihao Ding, Beining Wu, Jun Huang, arXiv

This one stands out because it combines privacy-oriented federated unlearning with freshness-aware optimization in mobile edge computing, making it more operationally relevant than purely theoretical privacy work.

🔗 https://arxiv.org/abs/2605.22589

Propagation-Consistent Wireless Environment Digital Twin Construction Under Sparse Measurements — Junjie Ai et al., arXiv

Interesting for wireless measurement and simulation workflows because it targets more realistic digital twin construction from sparse measurements rather than assuming dense sensing coverage.

🔗 https://arxiv.org/abs/2605.22361

🎓 MIT/Harvard Events This Week

• June 1 — Getting Started with Claude and Cowork @ 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/getting-started-claude-and-cowork-0?occ_id=0
• June 2 — Claude Code: Setup, Commands, and Context @ 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/claude-code-setup-commands-and-context-0?occ_id=0
• June 3 — Advanced Claude Code: Skills, MCPs, Hooks, and Multi-Agent Workflows @ 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
• June 4 — Claude Code: Putting it all together to develop course content and communicate research @ 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/event/claude-code-putting-it-all-together-develop-course-content-and-communicate-research?occ_id=0

⚠️ Source Issues

• The TNT calendar still returned stale February–April listings, so event picks were cross-checked on direct Harvard pages.
• arXiv API queries returned HTTP 429 rate limits during collection, so paper discovery fell back to live recent-category pages.
• IEEE Xplore and ACM did not surface strong fresh matches quickly in this run, so today’s research section leans on arXiv.

💡 Takeaway

Today’s pattern is operationalization: frontier AI, 6G research, and satellite connectivity are all moving closer to durable, deployable systems rather than isolated demos.

🧝‍♂️ Jarvis

Morning Digest — Thursday, May 21

Anthropic launches Claude 4

Anthropic introduced Claude Opus 4 and Claude Sonnet 4, positioning them as stronger models for coding and long-horizon agentic tasks. The broader signal is that leading labs are now competing on sustained tool use, reliability, and end-to-end task execution rather than just static benchmark performance.

Source: https://www.anthropic.com/news/claude-4

OpenAI says a general-purpose model cracked a classic Erdős problem

OpenAI says one of its reasoning models discovered a new family of constructions for the planar unit distance problem, challenging a decades-old assumption that the best solutions looked roughly like square grids. If the mathematical community validates the result, this is a notable example of AI producing original research insight rather than only assisting human problem-solving.

Source: https://x.com/OpenAI/status/2057176201782075690

Google pushes Android XR from concept to device ecosystem

At Google I/O, Google said Android XR devices are coming with Samsung, XREAL, Gentle Monster, and Warby Parker. That matters because it suggests AI assistants are being embedded into always-available wearable interfaces, which could become a more persistent computing surface than phones alone.

Source: https://blog.google/products/android/android-xr-platform-and-devices/

NVIDIA and Google Cloud say their joint developer community cleared 100,000 members

NVIDIA said more than 100,000 developers have joined its joint community with Google Cloud in a year, and highlighted new learning paths around JAX, GKE, and inference optimization. The practical takeaway is that the AI buildout is not just about chips anymore — it is about training large developer communities to deploy multi-agent and GPU-heavy systems efficiently.

Source: https://x.com/nvidia/status/2056891831820882402

Starlink says Direct to Cell is now available

Starlink’s support documentation now lists Direct to Cell texting as available, while voice and data remain marked as coming soon. For the satellite-and-wireless world, this is one of the clearest signs that direct-to-phone service is moving from promise to real user-facing availability.

Source: https://www.starlink.com/support/article/373ef6da-f2bf-5d14-9cb1-b6d25b59d1f1

T-Mobile, AT&T, and Verizon are forming a joint venture to tackle mobile dead zones

T-Mobile says the three major U.S. carriers will jointly fund and operate satellite-based coverage aimed at eliminating remaining dead zones. That is strategically important because it suggests satellite coverage is becoming shared national infrastructure rather than a single-carrier differentiator.

Source: https://www.t-mobile.com/news/un-carrier/t-mobile-att-verizon-jv-to-end-mobile-dead-zones

SpaceX targets Starship Flight 12 for today

SpaceX says Starship’s twelfth flight test is scheduled for Thursday, May 21, 2026, with the test window opening at 5:30 p.m. CT. Beyond launch spectacle, each successful iteration matters for the future economics of large-scale orbital systems, including communications and compute infrastructure in space.

Source: https://x.com/SpaceX/status/2057292990532481513

Research Radar

Communication Challenges in LEO Satellite Systems: Issues and Countermeasures for 6G and IoT Integration

Authors: Mathilde Lumineau, Jules Pénot, Hamsa Balakrishnan
Venue: arXiv
This paper maps the practical communications bottlenecks that show up when LEO systems meet future 6G and IoT workloads, especially around fading, latency, mobility, and integration complexity.

🔗 https://arxiv.org/abs/2605.11387

AI-Powered Satellite Networks: A Survey, Taxonomy, and Future Research Directions

Authors: Fazel Hossin, Tuan Minh Pham, Takahiro Hasegawa, Long B. Le
Venue: arXiv
A useful survey for understanding how AI is being positioned across satellite-network control, resource allocation, autonomy, and system optimization.

🔗 https://arxiv.org/abs/2605.12557

Resource Allocation for AI-Generated Services in 6G Networks

Authors: Yao Wang, Yao Teng, Shixiong Qin, Ning Zhang, Xiaohu Ge
Venue: arXiv
Worth watching because it reframes 6G resource allocation around the needs of AI-generated services, which is closer to the traffic mix future networks may actually face.

🔗 https://arxiv.org/abs/2605.12405

MIT/Harvard Events This Week

• May 21, 2026 — Claude Code: Putting it all together to develop course content and communicate research @ 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/generative-ai-events

• May 21–22, 2026 — Transforming Healthcare with AI @ MIT Sloan Executive Education, Cambridge
  Source: https://executive.mit.edu/on/demandware.static/-/Sites-master-catalog-msee/default/dw2018d431/brochures/THA%203.25.26.pdf

Source Issues

• TNT calendar still returned stale February–April listings, so this week’s event picks were cross-checked on direct Harvard/MIT pages.
• @ASTSpaceMobile returned no fresh usable posts.
• @jensenhuang surfaced only stale 2014-era posts, so it was excluded.
• IEEE Xplore and ACM did not yield strong fresh paper matches quickly, so today’s Research Radar leans on recent arXiv papers.

Takeaway

Today’s pattern is embodiment: frontier AI is escaping the chat box into mathematical discovery, wearables, developer ecosystems, and satellite-linked communications.

Morning Digest — Wednesday, May 20, 2026

OpenAI turns compute access into a product

OpenAI has launched Guaranteed Capacity, a new enterprise offering that lets customers commit to 1–3 year compute allocations for production systems, customer-facing applications, and agent workflows. The bigger signal is that frontier-model demand is now constrained enough that predictable access to compute has itself become a strategic product layer rather than a background infrastructure concern.

Source: https://openai.com/business/guaranteed-capacity/

OpenAI adds a public verification layer for AI-generated images

OpenAI is previewing a public image verification tool that checks for provenance signals including C2PA metadata and Google DeepMind SynthID watermarking. This matters because content provenance is finally becoming operational: not just a policy promise, but a usable verification workflow for platforms, journalists, and researchers.

Source: https://openai.com/index/advancing-content-provenance/

Anthropic lands a global KPMG alliance

Anthropic and KPMG announced a broad alliance that embeds Claude into KPMG’s Digital Gateway and expands access across a global workforce of more than 276,000 people. That makes this more than a branding partnership — it is a real signal that enterprise AI adoption is moving from exploratory pilots into deeply integrated work products inside large professional-services organizations.

Source: https://www.anthropic.com/news/anthropic-kpmg

Google pitches Gemini as a scientific co-worker

Google introduced Gemini for Science, a suite of experimental tools for hypothesis generation, literature understanding, and computational discovery. The notable shift is conceptual: Google is positioning multi-agent AI as a general-purpose research collaborator that can support the full scientific method, not just isolated search or summarization tasks.

Source: https://blog.google/innovation-and-ai/technology/research/gemini-for-science-io-2026/

Gemini app pushes further into proactive agent territory

Google says the Gemini app is becoming more agentic with proactive daily briefs, a redesigned interface, and Gemini Spark, a 24/7 assistant that can act under user direction. The strategic pattern is clear: consumer assistants are being redefined from reactive chat surfaces into persistent operators that stay engaged between prompts.

Source: https://blog.google/innovation-and-ai/products/gemini-app/next-evolution-gemini-app/

Ericsson and T-Mobile report live-network gains from AI-native RAN

Ericsson says commercial trials on live T-Mobile 5G Advanced traffic delivered about 10% higher spectral efficiency and up to 15% higher downlink throughput using an AI-native scheduler with link adaptation. For wireless researchers, this is one of the clearest recent field indicators that AI-native RAN is progressing from architecture rhetoric to measurable production-network benefit.

Source: https://www.ericsson.com/en/press-releases/6/2026/t-mobile-ericsson-ai-ran

Starlink leans into resilience as a core selling point

Starlink highlighted customer use during severe Midwest weather, emphasizing continuity through power outages and infrastructure disruption. Even though the example came through an X post, it reinforces a serious LEO-broadband theme: resilience and failover are becoming just as important to Starlink’s positioning as raw speed.

Source: https://x.com/Starlink/status/2056827466564739333

Research Radar

Movable Antenna-Aided Secure LEO Satellite Networks: Joint Antenna Position and Beamforming Optimization

Authors: Suhong Luo, Pan Tang, Jianhua Zhang, Ji Wang, Yixuan Li, Zihang Ding
Venue: arXiv
This paper studies how movable antennas and beamforming can be jointly optimized to improve secure LEO satellite links. It is directly relevant to the practical design of non-terrestrial networks that must remain robust under adversarial or interference-heavy conditions.

🔗 https://arxiv.org/abs/2605.18099

Delay-Aware Large-Small Model Collaboration over LEO Satellite Networks

Authors: Mingyu Guo, Wen Wu, Ying Wang, Songge Zhang, Liang Li
Venue: arXiv
This work frames orbital AI as a distributed systems problem, splitting tasks between constrained sensing satellites and better-provisioned computing satellites to reduce end-to-end delay. It is especially interesting for anyone thinking about in-network intelligence above Earth rather than only on the ground.

🔗 https://arxiv.org/abs/2605.04565

QoS Assurance Mechanism for 5G Network Slicing Based on the Deep Reinforcement Learning PPO Algorithm

Author: Qingyang Li
Venue: arXiv
This paper proposes a DRL-based QoS control loop for 5G slicing that jointly manages bandwidth, compute, and wireless resources. The useful angle is that it treats slice quality as a coupled systems problem instead of a radio-only one.

🔗 https://arxiv.org/abs/2605.03345

MIT/Harvard Events This Week

• May 20 — Advanced Claude Code: Skills, MCPs, Hooks, and Multi-Agent Workflows @ Harvard Bok Center, 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/generative-ai-events
• May 21 — Claude Code: Putting it all together to develop course content and communicate research @ Harvard Bok Center, 50 Church Street, Suite 374
  Source: https://bokcenter.harvard.edu/generative-ai-events
• May 21–22 — Transforming Healthcare with AI @ MIT Sloan Executive Education, Cambridge
  Source: https://executive.mit.edu/on/demandware.static/-/Sites-master-catalog-msee/default/dw2018d431/brochures/THA%203.25.26.pdf

Source Issues

• The TNT calendar page still surfaced stale February–April listings, so this week’s events were cross-checked on direct Harvard and MIT pages.
• AST SpaceMobile did not surface fresh usable updates during this run.
• Next G Alliance results were stale, mostly from 2023–2024, and were excluded.
• arXiv API queries hit rate limits, IEEE Xplore was robots-blocked in search, and ACM did not quickly surface clean fresh results in this environment, so today’s paper section leans on recent arXiv finds.

Takeaway

The strongest common thread today is persistence: AI is becoming infrastructure with reserved capacity, always-on agents, verifiable outputs, and measurable network-side intelligence rather than just better chat demos.