Russ

Author Archives: Russ

Hedge 316: AI Governance

Deploying AI for AI Ops, or even just for general use in your network, is very simple–but we often forget that these kinds of new technologies need to be governed. From privacy through cost, operators need to decide how to govern their AI deployments to control costs, ensure accuracy, measure productivity, and make certain these systems are being used effectively. Colin Cosgrove joins Russ and Tom to look at AI governance.
 

 
download

Worth Reading 081526


I am not a security researcher, and I have never been to Georgia. Yet within a few hours, using AI tools and nothing but public records, I was able to reconstruct the order in which 1.5 million ballots were cast in Georgia’s May 2026 primary–98.9% of the in-person ballots.

 


In profession after profession, GenAI is beginning to perform many of the tasks that traditionally served as training grounds for newcomers.

 


What’s the manifestation of the fallacy in the current case? Thinking that a system that is great at a certain kind of math problem is great at all math, great at science or even quite possibly great at everything.

 


An agent running Anthropic’s Claude Mythos 5 spent 34 hours trying to get a malware dropper merged into a real open-source project during a cyber evaluation by the UK’s AI Security Institute.

 


This week, DuckDuckGo is filing an amicus brief in the appeal of a federal court decision that Google unlawfully maintained a monopoly in the general search market in violation of the Sherman Antitrust Act.

Worth Reading 081026


There is an interesting fight within the leadership of U.S. AI companies. It is between the supporters of open-source AI and those driven to oppose open-source AI.

 


A Chinese man accused by the communist authorities of having committed “economic crimes” figured that he could evade the long reach of the country’s police state by hiding out in plain view: in a stadium packed with 60,000 people listening to a pop concert.

 


The J2SW Prefix Advertisement Checker, my 3rd application in the network series, checks the exact prefix against live routing data and reports the origin ASN seen by RIPE RIS collectors.

 


A growing number of drivers are pushing back against increasingly tech-heavy vehicles, saying they prefer the simplicity of a

 


The IETF does not launch rockets, or design spacecraft, or even work on radio systems and spectrum assignment, but it has been engaged in the standardisation of communications protocols used to communicate with these spacecraft. Here’s a few topics that were presented at IETF 126.

Worth Reading 080826


Would you work differently if you knew every keystroke was being saved, every website logged, every idle minute counted against you?

 


Everything I read about the AI industry leads me to think there will be an AI crash.

 


I designed this to help with different BGP communities and configuring them on your routers.

 


It’s commonly believed that when selecting between a number of routes for the same address prefix, a BGP speaker will select the route with the shortest AS Path.

 


There is an interesting fight within the leadership of U.S. AI companies. It is between the supporters of open-source AI and those driven to oppose open-source AI.

Hedge 315: Engineering Psychology

We often talk about “human infrastructure” in network engineering–but what does this mean, and how do we help engineers “produce” while helping them to flourish as individuals? Catherine Hicks joins Tom and Russ to discuss the psychology of software engineering, and how the lessons learned apply to network engineering.

download

You can find Catherine’s new book on the psychology of software engineering here.
If you would like to follow Catherine’s writing in this area, including the impact of AI on work, you can find her blog here.
You can connect with Catherine on LinkedIn.

Worth Reading 080126


The honest target is goodput, meaning requests per second that actually met your latency targets. Count anything else and you are counting cold plates.

 


I find it strange that as a software industry, we put in so much effort to make our systems efficient to design, build, and scale with DevOps practices, cloud native architectures, event-driven functions, optimized databases and storage, and resource-sipping Lean IT — then we threw it all out the window when Big AI came along.

 


If there is one universal law in internet infrastructure, it is that no good intention goes unpunished by the practical realities of legacy software maintenance.

 


This Article updates and expands on 2012 research on encryption and globalization, analyzing what the authors call “Round 3” of the Going Dark Debate: the current controversies over end-to-end encryption (E2EE).

 


Using our unique position on the Internet, we took an investigative look at one of the well-known mandatory attributes in BGP, the ORIGIN attribute.

Worth Reading 072826


Today’s Internet runs on a vast infrastructure of replicated data, and the difference between order and chaos lies in maintaining consistency through synchronization across the many distributed points where that data is published.

 


IPv6-only networks often still depend on IPv4 subnets and ARP. This article introduces an IETF proposal to eliminate both.

 


Traditional software engineering approaches reliability as a binary state of uptime and downtime. Modern ML pipelines render this paradigm obsolete.

 


Yet while the radios and protocols largely reuse terrestrial 5G designs, one small but crucial component has been left behind: The control plane timers that quietly govern how registration, mobility, and session procedures behave under real-world conditions.

 


Apparently, nearly all of them have reached the same conclusion. I don’t merely need AI occasionally. I need it waiting inside every search bar, messaging app, music player, and document reader I already use.

Hedge 313: Out of Band Management Networks

Most designers and operators default to carrying all traffic–including network management traffic–across the same paths as application data, or in band. Is this the right choice, or should operators seriously consider creating an out-of-band network just for network management? If you decide to build an out-of-band network, what are some of the considerations and lessons learned? Daryll Swer joins Tom and Russ to discuss.
 

 
download

Worth Reading 072226


Email authentication has quietly held the internet’s mail together for two decades, and it has just taken its biggest step forward yet.

 


A few months ago in Mercer County, Kentucky, more than 200 residents packed a planning and zoning commission meeting until every seat was filled, with citizens standing along the walls. Yard signs reading “NO DATA CENTER” lined farm roads from Harrodsburg to Burgin.

 


AI coding tools have changed the software development world in a few short years, but security risks and rising costs associated with AI-generated code mean organizations should consider their potential return on investment (ROI) before jumping into the deep end.

 


A recent PING episode noted that, for the Border Gateway Protocol (BGP) at least, ‘the shape of the curve’ appears to have changed.

 


Today’s Internet runs on a vast infrastructure of replicated data, and the difference between order and chaos lies in maintaining consistency through synchronization across the many distributed points where that data is published.

Worth Reading 071926


In fact, in Q1 2026, the Ethernet switch market grew faster than the overall server market did according to the latest statistics from IDC.

 


We are tempted to treat a capable language model as a knowledgeable colleague whose conclusions we can accept. It is more accurate, and more useful, to treat it as an untrusted component that produces plausible output which must be checked before it is relied upon.

 


In my view, MoQ occupies a middle ground between WebRTC (which is used for lots of video conferencing applications like Google Meet) and DASH (Dynamic Adaptive Streaming over HTTP) which powers most entertainment video streaming on the web.

 


Content consumption is detached from the website itself, as users rely on AI-driven systems to aggregate, summarize and contextualize information without visiting the original source. Instead of researching across multiple tabs, readers ask an AI system to do the work for them.

 


Shifting data centers from earth to space has become an alternative touted as solving concerns without generating new ones. In sun synchronous, low Earth orbit, orbiting data centers (“ODCs”) may have comparatively less environmental impact and lower operating costs.

Hedge 312: Keys and DNS

The entire technology world has, for decades, treated the IP address as a shorthand host identifier. This is clearly not the way IP was designed, but what are our other choices? In this episode of the Hedge, Scott Robohn joins Russ And Tom to discuss a recent paper arguing cryptographic keys should be the primary host identifier, and another article on the centrality of DNS to the Internet.

download

Worth Reading 071526


Reports of AI corporate implementation failures have continued to mount over the last year.

 


AI assistants now hand you a single, ready-made answer, and a harder question comes with it: who decides what we get to know, and what never makes it into the reply?

 


Vendors are trying to position “confidential computing” as the technical backbone of Europe’s sovereign cloud ambitions. But new research shows that a security protocol used to prove cryptographic trust in the system may have a fundamental architectural flaw.

 


The Internet should stop asking only who was in the room. It should ask who can bind the party bearing the loss.

 


America won’t beat China by banning AI. We’ll win by building the world’s best open models and letting innovation—not government—lead the way.

Worth Reading 071126


A unanimous Supreme Court reversed a $1 billion dollar jury award against a broadband provider in its March 2026 ruling in Cox Communications v. Sony Music.

 


Differential privacy (DP) data synthesizers are increasingly proposed to afford public release of sensitive information, offering theoretical guarantees for privacy (and, in some cases, utility), but limited empirical evidence of utility in practical settings.

 


LLMs are relentless data miners that train on unimaginably large text databases, looking for multi-dimensional statistical relationships among small chunks of text called tokens.

 


Regulatory filings and new business models suggest hyperscalers are shifting from simply building AI capacity to managing the enormous financial risks that come with it.

 


olicymakers are casting more and more problems as issues of cybersecurity. So reframed, wildly different policy issues, from misinformation, to child social media safety laws, to antitrust regulations, to alleged journalist misconduct, to anti-sex trafficking statutes become what this Article calls “cybersecuritized.

Hedge 311: The Dangers of AI

We are often told that if engineers who don’t go “all in” on AI will be left behind. While there are a few voices who argue that AI can be a dangerous tool, impacting not only the quality of our work, but the very quality of our thinking skills. Doug Smith, a critic of using AI for engineering, joins this episode of the Hedge to argue the contrary view of AI.

What are the dangers of relying on AI?

download

You can find Doug’s blog here.

Worth Reading 070126


A better, more nuanced understanding of tasks in recommender systems can help minimize user costs across the entire recommendation process.

 


The current estimate of the world’s population is 8.264 billion people, so the share price of SpaceX is currently at a phenomenal USD $261 per head.

 


The analysis from space monitoring firm LeoLabs, provided to Breaking Defense, found that from January 2021 to January 2025 China has abandoned 51 spent rocket bodies in LEO above 650 kilometers (about 404 miles) in altitude, more than doubling the number for the previous five years to bring the total to 96.

 


A big problem, Pauzauskie said, revolves around reproducibility. So far, labs haven’t been able to show that they can consistently cool semiconductors.

 


This article recounts the evolution of modern computing systems to provide an analysis of security risks and their evolution, with a past and present look at key cyber-defense innovations as well as a perspective on future cybersecurity hard problems.

Worth Reading 062926


It’s no longer just about your IP address or the specific endpoint you think you’re connecting to, it’s about your location and which intermediary services can most effectively handle your request.

 


Earlier this month, a German court ruled that Google is liable for its AI search summaries. Rejecting defenses like “users can check for themselves,” and that they generally know “that information generated with AI should not be blindly trusted,” the court held that the AI’s summaries are reflections of the company and “above all an expression of Google’s business activities.”

 


The distribution of Content Delivery Networks (CDN), cloud and content provider capacity across Internet Exchange Points (IXPs) provides a fascinating lens into the physical infrastructure of the Internet and public peering.

 


The memory market – by which we mean dynamic main memory as well as flash persistent memory – has been utterly and perhaps forever changed by the GenAI boom.

 


These days you could be excused by suspecting that the world has gone AI-mad, and if you were at the NANOG meeting your suspicions would’ve only been confirmed!

Worth Reading 062426


As the development of data centers in the U.S. faces intense criticism from local communities and legal action, project supporters are claiming “foreign influence” could be fueling the fire.

 


In this bingecast installment of the Mind Matters News podcast, host Robert J. Marks welcomes economics professor and author Gary Smith to discuss the hype around artificial intelligence and its impact on the market.

 


In the DNS name resolution space queries are free. To what extent do we see over-querying on the part of recursive resolvers in the DNS?

 


Over the past year, the Ethernet community has examined 400Gbps‑per‑lane signaling from many angles: AI network use cases, modulation options, channel limitations, and technology feasibility. Those discussions have been necessary, but they now need to converge into decisions that allow the industry to move forward.

 


For most of the internet era, distribution was scarce and content was abundant. Platforms that controlled distribution captured the majority of value.

1 2 3 … 169