Adapting to AI penetration testing orchestrators: The Modern Security Model

Over the last month, the number of Large Language Model (LLM) penetration testing and orchestration frameworks has exploded. Just a few weeks ago, ARTEX — an open source agentic penetration testing system that automates offensive security workflows — was the only framework available. In just the last three days, we have seen over 50 framework families attempt to scan Cloudflare customers globally. The number of Internet Protocol (IP) addresses seen running these frameworks has grown into the thousands and continues to increase daily. The proliferation of these frameworks has significantly changed the status quo in cyber defense: defenders are struggling to keep up with the volumetric nature of these autonomously guided attacks.

ARTEX was taken down by its creator Autumn-27 on October 8. (“The malicious exploitation of this tool runs entirely counter to its original intent,” they wrote in Chinese.) But in the month since, the ecosystem has exploded with new fragmented versions of it. There are now over 12,000 repositories, with several hundred unique forks containing lineage from the original repository. Additionally, hundreds of other similar agentic orchestration frameworks have been created since October 8. Daily requests of AI penetration testing (pentesting) and exploitation tools has grown by 30x since October 1. Continue reading

Introducing Spend Management: Spending limits and alerts for Cloudflare

Spend controls have been at the top of the list of most requested features from developers building on our developer platform for quite some time now. With more than 10 million developers now building on Cloudflare, this matters more than ever.

Although we’ve had usage notifications and budget alerts available to paid plans for a while, this doesn’t solve the problem completely; informing is not the same as controlling.

You can now set spending thresholds and choose to have Cloudflare pause activity when those limits are reached, thus avoiding unexpected bills from unplanned usage. Today, we're introducing Spend Management in Beta, for any pay-as-you-go (PayGo) customer, starting with Workers and Durable Objects.

Use Spend Management alongside AI Gateway spend limits to manage compute and AI inference costs across your application. AI Gateway already supports per-user and per-model budgets for Workers AI and other AI providers, applied to requests routed through your gateway.

How it works

You can start using Spend Management using the dashboard, cf CLI, or REST API.

In the dashboard, head over to Manage Account > Billing > Spend limits and enable the products for which you want to control spending.

When you enable controls on a Continue reading

Larangan Mesin Bensin 2035 dan Dampaknya di Eropa

Mengapa Kebijakan 2035 Menjadi Sorotan Dunia

Dalam beberapa tahun terakhir, salah satu topik paling panas di industri otomotif global adalah rencana penghentian penjualan mobil baru bermesin pembakaran internal di Eropa pada 2035. Kebijakan ini tidak hanya menjadi isu lingkungan, tetapi juga persoalan ekonomi, teknologi, tenaga kerja, dan strategi industri jangka panjang. Bagi sebagian pihak, langkah ini adalah terobosan penting untuk mempercepat transisi menuju transportasi rendah emisi. Bagi pihak lain, kebijakan tersebut menimbulkan kekhawatiran soal biaya, kesiapan infrastruktur, dan masa depan pabrik mobil tradisional.

Eropa memang dikenal sebagai wilayah yang agresif dalam mendorong target iklim. Dengan tekanan untuk menurunkan emisi karbon, sektor transportasi menjadi salah satu sasaran utama. Mobil penumpang menyumbang bagian besar dari polusi udara perkotaan dan emisi gas rumah kaca, sehingga peralihan ke kendaraan listrik dianggap sebagai solusi paling realistis dalam jangka menengah.

Apa yang Sebenarnya Dimaksud dengan Larangan Ini

Istilah “larangan mesin bensin” sering terdengar sangat absolut, tetapi kenyataannya lebih kompleks. Yang dibahas adalah penghentian penjualan mobil baru yang menghasilkan emisi karbon dari mesin pembakaran internal, bukan pelarangan total semua kendaraan bensin yang sudah beredar. Artinya, mobil lama masih bisa digunakan di banyak negara sesuai aturan yang berlaku, meskipun ruang geraknya bisa makin dibatasi melalui pajak, zona emisi Continue reading

Getting Started with Ansible and Arista

Getting Started with Ansible and Arista

This post is part of an ongoing series on Ansible, and you can find all the other posts in the series here.

I'm writing this post as part of my Ansible for Network Engineers series. But if you've just stumbled across this post without reading the previous parts, you should still be able to follow along. I won't go into detail on everything we've already covered, such as inventory, variables, loops, and handlers, but I'll point you to the relevant parts where it helps.

Earlier in the series, we built the OSPF topology below using Cisco routers, and we've also looked at configuring OSPF on Juniper. Now it's Arista's turn. The reason I keep coming back to this OSPF topology is that it's a great way to cover a lot of topics in one go. With six routers and three OSPF areas, we get to work with interfaces, IP addressing, OSPF areas, passive interfaces, and router IDs, and it gives us a good excuse to use a range of different modules and config commands along the way. It also makes it really easy to compare how each vendor does things differently, since the network itself stays exactly the same.

Getting Started with Ansible and Arista
arista Continue reading

Choosing QuePaxa, an unusual consensus algorithm

Cloudflare recently announced Meerkat, an experimental distributed consensus service designed to manage our internal global control plane. It is powered by a new consensus algorithm published by researchers Tennage & Băsescu et al. called QuePaxa (2023).

QuePaxa is an unusual choice; most consensus systems use well-established consensus algorithms like Raft, (Multi-)Paxos or Viewstamped Replication. These algorithms are well understood, have mature implementations, and usually work well.

Unfortunately, common consensus algorithms often suffer in wide-area networks with volatile network connectivity and latency, i.e., networks like the Internet, which is where Meerkat runs. These algorithms rely on a single leader process being available and having predictable latency. If the leader goes down, experiences higher latency than usual, or is disconnected, the entire consensus system slows down or becomes unavailable until the leader is replaced. This is unacceptable for Meerkat.

Meerkat’s global deployment environment led us to a newer consensus algorithm, QuePaxa, which promises better stability despite volatility, in large part by not relying on a leader to make progress. In this post we’ll explain why we chose QuePaxa and how the QuePaxa algorithm works.

What we’re building and why it’s unique

Meerkat is a distributed system in which replicas agree upon Continue reading

New Tool – BGP Prefix-List and Policy Generator (with a Policy Tester)

New on the Network Tools page: a BGP Prefix-List and Policy Generator. You describe a BGP import or export policy once, get the configuration for Junos, Cisco IOS / IOS-XE or Cisco IOS-XR, and then test it against a list of routes before it goes anywhere near a router.

What it builds

  • Prefix sets with ranges: exact, orlonger, longer, le, ge, ge N le M, prefix-length-range. Length checks are just ranges: “too specific” is 0.0.0.0/0 ge 25, “too short” is 0.0.0.0/0 le 7 (which also catches the default route). IPv6 works the same way.
  • Ordered terms. Each term matches on a prefix set, an AS-path condition (originated by, learned from neighbor AS, passes through, locally originated, contains a private ASN) and communities. Then it can set local-preference and MED, add, replace, delete or clear communities, prepend your AS, and accept, reject or continue to the next term.
  • Real platform output. Junos gets route-filter-list and policy-statement terms. IOS gets prefix-lists, AS-path ACLs, community lists and a route-map with one sequence per term. IOS-XR gets prefix-sets, as-path-sets, community-sets and a route-policy that ends in an explicit done or drop.
  • Continue reading

AI agents belong on your Kubernetes platform

In Kubernetes, pods get a new IP every time they restart, so an IP can’t tell you which workload it is. Platform teams learned to identify workloads by labels and identity instead. AI agents are now teaching them a second lesson: an agent’s code can’t tell you what it will do.

I believe AI agents belong on the Kubernetes platform. Not beside it, not around it, and not left to each team to secure alone. On it: identified, governed and contained like every other production workload.

I’ve spent the last decade working with the platform teams behind some of the largest Kubernetes environments in the world: banks, telecom carriers, SaaS companies and insurers. I watched them take containers from a side experiment to the default way their companies ship software. The best of them did it without slowing anyone down, and without giving up control.

This year I’m watching the same story start again with a new workload. AI agents are arriving on those platforms faster than anyone can govern them. In an April 2026 Cloud Security Alliance survey of 418 IT and security professionals, 82% said they had found at least one AI agent or autonomous workflow in the Continue reading

Performance evolution of route lookup on Linux

In previous articles, I explained how the Linux kernel implements IPv4 and IPv6 routing tables, and I showed the performance improvement of route lookups from Linux 2.6.32 to 4.16 for IPv4—including the impact of the Meltdown and Spectre mitigations—as well as from Linux 2.6.39 to 4.18 for IPv6. This article updates the numbers through Linux 7.3 and investigates the perpetrator responsible for the performance regressions! 🕵️‍♂️

I ran the following benchmarks on an AMD EPYC 8024P, a Zen 4c CPU, rented from Scaleway, running at 2.4 GHz with CPU boost disabled. I compiled the kernels with GCC 10.2: this version can compile older kernels as well as the current ones.

IPv6

Performance evolution of IPv6 route lookups from Linux 4.14 to 7.3
Performance evolution of IPv6 route lookups. Versions with an asterisk have SRSO mitigation disabled.

I tested four scenarios:

  • a recent full view with about 260,000 prefixes,
  • a full view from 2017 with about 40,000 prefixes,
  • a set of 40,000 /48 prefixes, such that 2 lookups out of 3 match, and
  • a set of 40,000 random /128 prefixes, such that 1 lookup out of 400 matches.

In Linux 4.15, Wei Wang improved the lookup performance Continue reading

Securing AI Agents Requires More Than a Gateway

Over the past year, I’ve spent a lot of time talking with engineering and platform teams about how they’re deploying AI agents. The conversations are changing. Six months ago, most were focused on getting agents into production. Now, increasingly, the question is how to secure them once they’re there.

It’s a familiar transition. We’ve seen it with containers, Kubernetes, microservices, and cloud infrastructure. New technology gets adopted, architectures evolve, and security has to catch up with how systems actually operate.

AI agents introduce some interesting challenges, though. Unlike traditional applications, agents make decisions about what to do next. They interact with models, invoke tools, communicate with other agents, access data, and execute code. A single user request can trigger a sequence of actions across multiple systems, sometimes with very little human involvement.

That changes the security problem.

An agent can be properly authenticated, operate within its assigned permissions, and still be led into doing something nobody intended.

It’s one of the reasons I believe securing agentic AI requires a defense-in-depth architecture, with controls operating at both the application and runtime layers.

The gateway is an important part of the architecture

AI gateways have become a natural place to introduce security Continue reading

Calico on Red Hat OpenShift 5: One Network and Security Model for Containers, VMs and AI

We’re pleased to announce that Calico is supported on Red Hat OpenShift 5, the latest version of the industry’s leading hybrid cloud application platform powered by Kubernetes. Tested and validated to run on Red Hat OpenShift 5, Calico delivers high-performance networking and zero-trust network policy for containers, virtual machines on Red Hat OpenShift Virtualization, and AI workloads, with one consistent policy model across the cluster.

Built on a long-standing collaboration

Calico has been a supported networking option for Red Hat OpenShift since OpenShift 4, and the Tigera Operator is listed in the Red Hat Ecosystem Catalog. Red Hat OpenShift customers can install Calico through the Tigera Operator on a new cluster, or move existing clusters from OVN-Kubernetes to Calico using Tigera’s documented migration path.

What Calico brings to Red Hat OpenShift 5

  • High-performance networking, your choice of data plane. Calico runs on eBPF, iptables or nftables, so teams can match performance and operational needs cluster by cluster.
  • Zero-trust network policy. Calico enforces Kubernetes NetworkPolicy and extends it with Calico’s own policy model, including cluster-wide rules, so default-deny and microsegmentation apply consistently.
  • Policy you can test before you enforce it. Staged policies show the effect of a rule on live traffic Continue reading

Protect-RE Filter Generator Update – Mist, TACACS+, DHCP and Custom Rules

The Juniper Protect-RE Filter Generator on the Network Tools page picked up a round of additions. The trigger was simple: we compared it against the protect_re filter a Mist-managed EX4400 actually carries, and found a handful of things a real-world filter needs that the generator didn’t cover yet.

What’s new

  • Mist-Managed Device. A Mist switch opens its connections outward, so the filter only has to accept the return traffic. Three optional terms cover it: outbound SSH (TCP source-port 2200), the Mist cloud over HTTPS (443), and NAC RadSec (2083). Each is TCP with tcp-established, so a fresh inbound SYN doesn’t match.
  • TACACS+. Return traffic from your TACACS+ servers (TCP source-port 49, tcp-established), with the server list pulled live from system tacplus-server using the same apply-path trick the BGP, NTP and RADIUS terms use. It’s TCP-only on purpose: TACACS+ doesn’t run over UDP, and tcp-established can’t match UDP anyway, so a combined TCP/UDP term can only ever match TCP.
  • DHCP and DHCPv6. A client option for routers that get their management address by DHCP, and a server/relay option for routers that hand them out. The server/relay term is placed ahead of anti-spoof, because a client’s first packet comes from Continue reading
1 2 3 … 3,921