Fixing Ghost Drops: How eBPF Rescued IPv6 Telemetry


A customer complains that they aren't getting IPFIX flow data from a router.

Use socat to check that IPFIX is being received (IANA assigned port for IPFIX is 4739):

socat -b 0 -dd -u UDP6-RECV:4739 - 2>&1
Output demonstrates that at least some IPFIX messages can be received when listening on port 4739. 
2026/05/15 22:46:32 socat[108419] N using stdout for writing
2026/05/15 22:46:32 socat[108419] N starting data transfer loop with FDs [5,5] and [1,1]
2026/05/15 22:46:33 socat[108419] N received packet with 0 bytes from AF=10 [fec0:0000:0000:0000:0001:000c:2744:69f1]:50978
2026/05/15 22:46:33 socat[108419] N received packet with 0 bytes from AF=10 [fec0:0000:0000:0000:0001:000c:2744:69f1]:50978
Use tcpdump to check for IPFIX packets. This gives visibility into packets before the host network stack, so you can see packets before they are dropped by host network stack or host firewall 
tcpdump -i enp0s3 -n udp port 4739
The output shows that IPFIX datagrams are being received from a second source, fec0::1:c:2744:69f0, but they aren't showing up in the socat output, so the Linux kernel must be dropping them for some reason. 
dropped privs to tcpdump
tcpdump: verbose output suppressed, use -v[v]... for full protocol decode
listening on enp0s3, link-type EN10MB (Ethernet), snapshot length 262144 bytes
21:09:57.217821  Continue reading

Announcing Claude Managed Agents on Cloudflare

Cloudflare and Anthropic have collaborated to integrate Claude Managed Agents with Cloudflare Sandboxes. Our new integration gives you more control over your agent sandboxes, secures connections to private services, and improves observability.

In the past year, Cloudflare’s Developer Platform has expanded to give more developers the tools they need to run agents at scale. This includes:

  • Sandboxes for full stateful Linux microVMs at scale

  • Agents SDK, providing simple and customizable agent framework

  • Browser Run, which gives agents fully programmable and observable browsers

  • Dynamic Workers, allowing for dynamic sandboxed code execution at massive scale

Our goal is to make Cloudflare the simplest, most secure, and most programmable cloud for agents.

Integrating with Claude Managed Agents is another step in this direction. You can run your agent loop on the Claude Platform, while using Cloudflare to execute code, secure connections, and run custom tool calls.

To get going in just minutes, we’ve created a default deployment template that gives you the following:

  • Enhanced security - Run all agent traffic through customizable proxies. This allows you to securely inject credentials, prevent data exfiltration, and better observe how your agents interact with the outside world.

  • Sandbox control and observability - Get Continue reading

Everything in C is undefined behavior

If he had been a programmer, Cardinal Richelieu would have said “Give me six lines written by the hand of the most expert C programmer in the world, and I will find enough in them to trigger undefined behavior”.

Nobody can write correct C, or C++. And I say that as someone who’s written C and C++ on an almost daily basis for about 30 years. I listen to C++ podcasts. I watch C++ conference talks. I enjoy reading and writing C++.

C++ has served us well, but it’s 2026, and the environment of 1985 (C++) or 1972 (C) is not the environment of today.

I’m definitely not the first to say this. I remember reading a post by someone prominent about a decade ago saying that a good case can be made that use of C++ is a SOX violation. And while I was not onboard with the rest of their rant (nor their confusion about “its” vs “it’s”), I never disagreed about that point.

With time I found it to be more and more true. WAY more things are undefined behavior (UB) than you’d expect.

Everyone knows that double-free, use after free, accessing outside the bounds of an Continue reading

Worth Reading 051826


We’ve all been in that meeting where someone pulls up a chart and says, “Our AI product boosted conversion by 15%.” Everyone nods. Nobody dares to ask: “What if conversions had risen anyway?”

 


The proposed repair is Running-Code Primacy: the number-resource layer should be interpreted only by reference to the minimum technical function running networks require—uniqueness, interoperability, proof of control, routing-adjacent security, and locally verifiable state.

 


You already know IPv6 is overdue. You’ve known for years. You’ve probably sat in a meeting where you laid out the case — address exhaustion, rising costs, growth constraints — and watched leadership nod politely before approving the budget for another batch of leased IPv4 addresses.

 


A key question was whether this reflected a breakthrough specific to one model, or part of a broader trend. Results from an early checkpoint of GPT-5.5 suggest the latter: a second model, from a different developer, now reaches a similar level of performance on our cyber evaluations.

 


The most mature U.S. small modular nuclear reactor vendor — NuScale Power — and a politically connected firm planning to build perhaps the largest reactor project in the U.S. to power an Continue reading

NB575: AI Multipath Protocol Goes to Open Compute Project; Cisco Shrinks Workforce as Income Swells

Take a Network Break! In this week’s Red Alert we suggest an audit of your Azure environment after Microsoft says it patched four critical vulnerabilities. On the news front, Nvidia has brought the Multipath Reliable Connection (MCR) protocol to the Open Compute Project, AT&T rolls out quantum-resistant SD-WAN services, and HPE introduces new Wi-Fi automation... Read more »

Project Glasswing: what Mythos showed us

For the last few months, we've been testing a range of security-focused LLMs on our own infrastructure. These LLMs help identify potential vulnerabilities in our own systems, so we can fix them – and they also show us what attackers are going to be able to do with the latest models.

None of these LLMs has captured more attention than Mythos Preview, from Anthropic. A few weeks ago, we were invited to use Mythos Preview as part of Project Glasswing. We soon pointed it at more than fifty of our own repositories – to see what it would find, and to see how it works.

This post shares what we observed, what the models did well and what they didn't, and how the architecture and process around them needs to change, so they can be used at scale.

What changed with Mythos Preview

Mythos Preview is a real step forward, and it's worth saying that plainly before getting into anything else. We've been running models against our code for a while now, and the jump from what was possible with previous general-purpose frontier models to what Mythos Preview does today is not just a refinement of what came before.

It's Continue reading

Goodbye, Ubuntu 20.04 (netlab 26.05)

netlab release 26.05 is out. I’ll write about its highlights tomorrow; today, I want to focus on one of its breaking changes: netlab no longer works with Python 3.8 (which reached end-of-life in October 2024), so you can no longer install it on a vanilla Ubuntu 20.04 (which reached end of standard support a year ago).

We wanted to get rid of old Python versions for ages, but never did because Ubuntu 20.04 shipped with Python 3.8, and many netlab early adopters installed it on Ubuntu 20.04 (and the last thing a networking engineer wants is wasting time with upgrades, right?).

Read more …

Automatic SSL with Caddy, Cloudflare and Let’s Encrypt

Automatic SSL with Caddy, Cloudflare and Let's Encrypt

I run a lot of applications and services in my homelab, and they usually look something like http://10.10.10.10:8000, http://10.10.10.11:8006, and so on. It works, but it comes with a few annoyances. Browsers constantly complain you about insecure connections, and remembering which IP and port maps to which service gets old fast.

The ideal setup is something like https://netbox-lab.yourdomain.com - a proper domain name with a valid SSL certificate and no port numbers to remember. In this post, I'll show you how to get there using Caddy, Cloudflare, and Let's Encrypt with very minimal effort.

Overview

To make this work, we need the following.

  • A Cloudflare account with a domain (You don't necessarily need Cloudflare. Other DNS providers should work the same way, but this post uses Cloudflare as the example.)
  • An LXC or VM to run Caddy
  • A Cloudflare API token with DNS edit permissions
  • Let's Encrypt Certificates are automatically managed by Caddy

Cloudflare Setup

I already have a domain that I use in my homelab, so I name my services like app.yourdomain.com. I purchased this domain through Cloudflare and manage it there.

To generate an API Continue reading

Using NetBox as a Dynamic Inventory Source for Nornir

Using NetBox as a Dynamic Inventory Source for Nornir

If you have used Nornir before, you already know that the most common way to define your inventory is through YAML files. You list your hosts, groups, and defaults in separate files, and Nornir reads them at runtime. This works fine for small labs or quick scripts, but as your network grows, maintaining those files by hand becomes a problem. Every time you add a device, change an IP, or update a site, you have to remember to update your inventory files as well. It is easy to get out of sync.

This is where a source of truth tool like NetBox can help. NetBox is an open source network source of truth. It is a place to document your devices, IP addresses, sites, roles and pretty much anything about your network. Instead of maintaining a separate YAML inventory alongside NetBox, you can pull your inventory directly from NetBox at runtime. That way, Nornir always works with up to date data, and you only have to manage one source of truth.

In this post, we will cover how to use NetBox as a dynamic inventory source for Nornir using the nornir_netbox plugin. If you are new to Nornir, we have Continue reading

Worth Reading 051526


Given the trend of using generative AI tools like ChatGPT, Gemini, Copilot, and Claude for software development, many companies have decided that developers must use GenAI to succeed. I strongly disagree.

 


Here, through a series of randomized controlled trials on human-AI interactions (N = 1,222), we provide causal evidence for two key consequences of AI assistance: reduced persistence and impairment of unassisted performance. Across a variety of tasks, including mathematical reasoning and reading comprehension, we find that although AI assistance improves performance in the short-term, people perform significantly worse without AI and are more likely to give up.

 


Last month, market research company, Gartner, said that AI companies need close to “$2 trillion per year in revenue by 2029”, token consumption of between 50,000 and 100,000 times its current rate by 2030, and “a 10% profit margin per token.” With huge losses and small revenues, it is not likely that AI companies will achieve these goals on time.

 


He said that about 20% of all network traffic today, about 80 exabytes, comes from machine-to-machine traffic, and that alone is big news. Nokia is betting its future growth will come from meeting this growing demand.

 


For Continue reading

Lab: EVPN Asymmetric IRB with Anycast Gateways

I postponed the discussion of ARP issues with EVPN anycast gateways to keep yesterday’s blog post reasonably short. If you’re impatient and want to try that out, I have just the right lab exercise for you; you’ll have to extend VLANs into end-to-end MAC-VRF instances and add IRB and anycast gateways:

Explore the lab exercise

You can run the lab on your own netlab-enabled infrastructure (more details), but also within a free GitHub Codespace or even on your Apple-silicon Mac (installation, using Arista cEOS container, using VXLAN/EVPN labs).

Cisco AUTOCOR Passed

Yesterday I took and passed the Cisco AUTOCOR (previously DEVCOR) exam which is the core exam for CCNP Automation. That means I need a specialist exam to become CCNP Automation certified. It also means I’m qualified to sit the CCIE Automation lab.

What did I think of the exam?

As with any exam, there is good and bad. I’ll start with the good.

The exam aligned well with the blueprint. I didn’t feel there were any real surprises or questions on items that weren’t part of the blueprint.

There wasn’t a lot of trivia. No memorization of specific API endpoints or anything like that.

The exam experience was fine. I took it in a testing facility, which I prefer, and I had no issues. I was provided with earplugs which was nice to stay focused although this is a small facility and there was only one other candidate.

I liked the different types of questions. You have your standard multiple choice, single answer and multiple choice, multiple answer, but also fill in the blanks, and lablets. It’s nice that there is quite a bit of code in the exam, it is an automation exam after all. I also think it’s Continue reading

KubeVirt Live Migration Done Right: What it Takes to Run VMs on Kubernetes

Running VMs in Kubernetes sounds like a crazy workaround for avoiding vendor lock-in, and standardizing legacy applications and newer containerized workloads on one control plane with one set of security policies to govern them all. It is, however, a rapidly growing pattern, and KubeVirt live migration — moving running VMs between nodes without downtime — is increasingly central to platform engineering use cases that require full VMs, like on-demand CI/CD pipelines.

KubeVirt is gaining traction as a way to bring VMs into Kubernetes as first-class workloads, managed with the same tools and primitives that platform teams already use for containers. It has, however, introduced some unique challenges.

Here’s the uncomfortable truth about that migration: compute and storage are the easy parts. Networking is where migrations stall, roadblock multiple, and platform teams start questioning whether KubeVirt was the right call in the first place.

If your VMs have no fixed IP dependencies, no VLAN memberships, and no upstream firewall rules scoped to specific subnets, you can migrate them into Kubernetes without losing sleep over the networking layer. If you’re running hundreds or thousands of VMs with IP addresses hardcoded into application configs, DNS entries, and firewall ACLs — and you need Continue reading

The AI Agent Accountability Crisis: Why Governance Isn’t Keeping Up With Deployment

Every enterprise is building AI agents. Marketing has one summarizing campaign performance. Engineering has one triaging incidents. Customer support has one resolving tickets. Finance has one processing invoices. Each was built by a different team, using a different framework, with different assumptions about security.

Now those agents are talking to each other through agent-to-agent (A2A) communication. The incident-triage agent calls the customer-support agent to check affected accounts. The invoice agent calls an external payment API. The marketing agent queries a data warehouse with customer records.

When something goes wrong (and at this scale of deployment, it will), can you answer:

  • Who authorized the action?
  • What policy permitted it?
  • What was the full chain of events?

If you can’t, you have an accountability gap.

This is part one of a five-part series on AI agent accountability for engineering and security leaders. We’ll work through the gap between agent deployment and governance, the diagnostic framework that exposes it, why your existing tools won’t close it, and the principles you’ll need to evaluate any solution that claims it can.

What is AI agent accountability?

AI agent accountability is the ability to trace, prove, and audit every action an AI agent takes. This includes Continue reading

1 … 39 40 41 42 43 … 3,915