Come To Think of It...

Every vendor on every panel right now is shouting the word "agentic." It has become the industry's favorite linguistic garnish, but there is a massive difference between a chatbot that can summarize a PDF and an actual agentic system that can govern a modern infrastructure. If we stop treating Governance, Risk, and Compliance (GRC) like a static filing cabinet and start treating it like a fluid, real-time system, the entire paradigm shifts. The fundamental problem is that our infrastructure has already gone agentic. Cloud is elastic, identity is fluid, and CI/CD pipelines are ephemeral. Yet, many compliance programs are still trying to govern these high-speed, non-deterministic environments using point-in-time snapshots and static checkboxes. You can't catch a real-time drift with a quarterly audit. To bridge this gap, an agent needs three things: autonomy to act when conditions change, the ability to execute multi-step workflows, and—most importantly—context. Th...

The problem with AI coding agents today isn't that they can't write code — it's that they have no idea what the rest of the system looks like. They see the file you opened, the diff you're working on, maybe the terminal output. But they don't know which service depends on the module you're changing, whether a similar vulnerability was flagged in another repo last month, or who reviewed comparable changes six months ago. That context lives in your DevSecOps platform, and getting it to a coding agent has meant custom scripts or copy-pasting between browser tabs. GitLab's new Orbit project, now in public beta, tries to close that gap with a live queryable graph that maps every relationship between groups, projects, merge requests, pipelines, vulnerabilities, and source code — and it just announced a direct integration with Google Antigravity agents. The integration itself is straightforward: agents install Orbit from the Antigravity MCP Store, query the graph...

Last month, Anthropic announced it was splitting Claude subscription usage in two: interactive chat and terminal use would stay in your normal subscription pool, while Agent SDK calls through third-party tools like Zed would move to a separate capped monthly credit. The numbers were stark — Pro users would get a $20 credit, Max 5x users $100, Max 20x users $200. For power users running agents through ACP in editors like Zed, the math was brutal. The team behind Zed calculated that Claude subscriptions had been subsidizing agent usage at roughly 15 to 30 times the equivalent API rate. The new credits would reset those costs to full API pricing, which the Zed team warned would be "a major cost increase" for anyone using agents heavily. Then, on the very day the change was supposed to kick in — June 15 — Anthropic hit pause. The company updated its billing support page with a blunt note: "For now, nothing has changed." The monthly credit is not available. Agent SDK u...

There is a new vulnerability in Microsoft 365 Copilot Enterprise called SearchLeak, and on its own it is the kind of finding that generates a CVE, a Microsoft blog post, and a few security team ticket updates. But the architecture of the attack tells a bigger story about what happens when prompt injection meets the old bug classes that security teams have been wrestling with for decades. SearchLeak chains three separate weaknesses into a silent data exfiltration pipeline. The first is a parameter-to-prompt injection: the search q parameter in Copilot Enterprise Search passes user input directly into the AI engine, treating it as both a search query and executable instructions. Unlike the regular Copilot chat feature, Enterprise Search is designed to pull from company data — emails, SharePoint files, OneDrive documents — so an injected prompt can surface anything the victim user has access to. The second weakness is an HTML rendering race condition: during Copilot's streaming res...

The Arch User Repository has always operated on a simple premise: if you trust the maintainer, you trust the package. That assumption has been quietly broken by what researchers are calling "Atomic Arch," one of the largest AUR supply chain attacks to date. Starting around June 11, threat actors began hijacking hundreds of orphaned packages — abandoned projects whose maintainers had simply stopped showing up — and modifying their PKGBUILD files to install a malicious npm package called atomic-lockfile during installation. Sonatype estimates the campaign may now affect as many as 1,500 packages across multiple waves. The trick isn't in the package itself but in what it pulls in. The PKGBUILDs were modified to run a post-install script that invokes npm install atomic-lockfile , and that npm package bundles a native Linux ELF binary with two main payloads. The first is an eBPF rootkit that hooks getdents64() to hide processes, files, and network interfaces from the user....

There is a specific kind of supply-chain attack that never quite gets old: you don't need to create trust from scratch, you just wait for someone to abandon their project and then step into their shoes. That is exactly what happened to the Arch User Repository, where a threat actor has quietly adopted over 400 orphaned packages, modified their build scripts to pull in a malicious npm package called atomic-lockfile , and sent a credential-stealing rootkit down to whoever happens to run yay -S on those packages. The story has two waves. The first, reported by Sonatype on June 11, targeted orphaned AUR projects whose maintainers had gone quiet — the attacker simply took ownership and changed the PKGBUILD to run npm install atomic-lockfile during the package install. That npm package bundled a native Linux binary with an eBPF rootkit (using libbpf to load, attach, and pin BPF maps) that can hide processes, files, and network interfaces, plus an infostealer that targets GitHub crede...

The Arch User Repository is supposed to be a community-curated heaven for power users — packages you can't find in the official repos, bleeding-edge versions, niche utilities, the stuff that makes Arch Arch. But it's never been vetted, and that's the thing most Arch users implicitly accept when they run yay or paru without a second thought. This week, that tradeoff stopped being theoretical. Over 400 packages across the AUR have been compromised to deliver a Linux rootkit and infostealer, according to a report from the Independent Federated Intelligence Network (IFIN). The threat actor spoofed a trusted publisher, changed package maintainers without anyone noticing, and embedded preinstall scripts that pull down a malicious npm package called atomic-lockfile . From there, a compiled ELF payload named deps — which independent researcher Whanos described as "a credential stealer with optional root-only eBPF rootkit capabilities" — takes over. What makes this at...

Linux kernel bugs usually require some cleverness to exploit. CVE-2026-23111, found in the nf_tables firewall subsystem, requires exactly one character: ! . Researchers at Exodus Intelligence discovered that placing a single exclamation point in an nftables verdict map definition flips the logic of a chain deletion check, which then lets an unprivileged user decrement the chain reference counter an arbitrary number of times. Delete the chain when its counter says zero, and you get a use-after-free — objects still point to memory that no longer belongs to them. From there, leak the kernel base address, hijack control flow, and you're root. The researchers clocked a >99% stability rate on an idle system. What makes this one worth a closer look isn't the severity — it's the elegance. A use-after-free in nf_tables is nasty, sure. But the path from a single punctuation mark to full root, without needing a second vulnerability, is unusually clean. The bug was patched upstrea...

Microsoft's June Patch Tuesday was a record-breaker. Nearly 200 CVEs across Windows OS and supported software, nearly three dozen rated "critical," and exploit code for at least three of those weaknesses already sitting on GitHub. But the headline isn't just the numbers — it's what Microsoft and the security community are saying about why the numbers are climbing. \n\n Microsoft's engineers and the wider security community are increasingly leaning on AI tools to find bugs. "Some surveys put AI usage among security professionals generally at 90%, so it's unsurprising that this volume of patches may become the norm," said Satnam Narang, a senior research engineer at Tenable. Pandora's proverbial box is open: as better models get better at pattern-matching through massive codebases, the yield of discovered vulnerabilities goes up. That means more patches, more maintenance windows, and more sysadmins refreshing Windows Update on a Tuesday mornin...

It is a peculiar, almost poetic reality of systems engineering that the difference between a stable kernel and a full-scale security breach can sometimes be a single, errant character. This week, the tech world caught wind of CVE-2026-23111—a high-severity vulnerability in the Linux kernel's nf_tables subsystem. The culprit? A single mis-issued exclamation point in the code. It sounds like a joke, but in the context of use-after-free (UAF) bugs, it's a textbook example of how human error in logic flows translates directly into memory corruption. \n\n The nf_tables subsystem is the backbone of modern Linux packet filtering, replacing the older iptables infrastructure. It manages firewall rules by determining 'verdicts'—the actions taken when a packet matches a rule. The bug occurs because the deletion of these verdict maps can be manipulated. By exploiting the way catchall elements (the wildcards of the set) are deactivated and how reference counters are decremented, ...

There is a specific kind of operational friction that only platform engineers truly understand: the 'agent update' deadlock. You need to update a monitoring tool, a logging driver, or a security scanner across a fleet of thousands of instances. But because that agent is technically part of the task definition, you can't just 'push' the update without coordinating with every single application team that shares those instances. It’s a classic case of the infrastructure's needs being held hostage by the application's lifecycle. Amazon just announced a way out of this with managed daemon support for ECS Managed Instances. The core shift here is the decoupling of the daemon's lifecycle from the application's. By moving daemons into their own managed construct, platform teams can now independently deploy, update, and even enforce specific versions of monitoring or tracing agents without touching a single line of the application's task definition. Thi...

Cloudflare has quietly done one of the more interesting open-source acquisitions in recent memory. VoidZero—the company behind Vite, Vitest, Rolldown, Oxc, and Vite+—has been acquired in what both sides are calling an "acqui-hire," meaning every team member joins Cloudflare. On the surface it sounds like the worst-case scenario for a beloved tool: the company behind your build tool gets swallowed by a CDN with a developer platform. But Evan You and Cloudflare's Steve Faulkner put out a joint blog post making three claims worth actually tracking: Vite stays MIT-licensed and vendor-agnostic, the roadmap stays community-driven, and Cloudflare is committing a $1 million ecosystem fund administered by the Vite core team to pay maintainers and contributors. The real story here isn't the acquisition itself—it's the precedent. Earlier this year Astro joined Cloudflare under the same arrangement, and it's still open source, still deploys anywhere, and the team is stil...

Cloudflare has acquired VoidZero, the small team behind Vite, the JavaScript build tool that nearly every modern framework now depends on. Evan You, creator of both Vue.js and Vite, is moving into Cloudflare, along with the rest of the VoidZero crew. The catch, as always with "acqui-hires", is what happens to the open-source projects: Vite, Vitest, Rolldown, Oxc, and Vite+ will stay open-source, vendor-agnostic, and community-driven. Cloudflare is also pledging $1 million to a Vite ecosystem fund administered by the core team. On the surface this reads like a generous move — Cloudflare, the DDoS-protection-and-DNS company, is buying the team that built the single most universally-accepted frontend toolchain in the JavaScript ecosystem and promising not to change anything. The detail that matters, though, is the Vite Environment API. Worked on jointly since 2024, it lets Vite run server code in something other than Node.js during development — which means running server code...

It's not often that a CISA update feels like a genuine "check your configs" moment, but the latest entries in the Known Exploited Vulnerabilities (KEV) catalog are hard to ignore. We're looking at two distinct flavors of trouble: a high-severity integer overflow in the Android framework (CVE-2025-48595) and a privilege escalation flaw in the Linux kernel (CVE-2022-0492) that is particularly nasty for anyone running containerized workloads. The Android issue is a classic: an integer overflow in the framework that can grant elevated privileges with zero user interaction. It's hitting Android 14 through 16, which means it's sitting right in the middle of the current mobile landscape. But for those of us more concerned with the server room or the cloud, the Linux kernel flaw is the real sleeper. It targets the cgroup_release_agent_write() function in the cgroups v1 subsystem. If you're running containers that haven't migrated to cgroups v2, an attacker...

We've spent a lot of time talking about the difference between "chatbots" and "agents," but seeing the tools actually catch up to the hype is a different experience entirely. Anthropic's latest heavy hitter, Claude Opus 4.8, just landed on AWS via Amazon Bedrock and the Claude Platform, and it feels less like a minor version bump and more like a fundamental shift in how these models handle long-running tasks. The real meat here isn't just "more reasoning"—it's the focus on agentic coding and extended autonomous execution. According to the recent AWS roundup, Opus 4.8 is specifically tuned to sustain longer sessions, recover from errors mid-task, and synthesize information across massive contexts. For anyone building automation workflows, this is the difference between a model that gives up when an API call fails and one that actually tries to debug and retry. It's moving us closer to the reality where we aren't just prompting a m...

A newly disclosed vulnerability dubbed CIFSwitch is forcing Linux administrators to take a hard look at their file sharing configurations. The flaw, which reportedly grants root access across multiple distributions, exploits a previously unnoticed interaction within how certain Linux environments handle specific network file system requests. For platform engineering teams and self-hosters alike, it’s a stark reminder that foundational infrastructure components—the ones that sit quietly in the background for years—remain prime targets for critical escalation paths. While the technical specifics center on how the system processes CIFS/SMB mounts under specific conditions, the operational reality is more pressing. Many organizations treat internal file shares as low-risk zones, often deploying them with default settings behind a firewall. CIFSwitch demonstrates that when a service requires elevated privileges to mount and manage networked file systems, any edge case in its input validati...

It’s a subtle but significant admission of defeat from Microsoft. At Build 2026, they didn't just announce a new feature; they announced a surrender to the reality of modern developer workflows. By releasing 'Coreutils for Windows'—built on the Rust-based uutils project—they are effectively admitting that the native Windows command-line experience has been a friction point for far too long. The technical implementation is actually quite clever, if a bit unorthodox. Instead of a massive overhaul of the kernel or even just another layer of WSL, they've bundled the functionality into a single coreutils.exe binary and used NTFS hardlinks to create individual command wrappers like ls.exe , cat.exe , and rm.exe . This gives the illusion of a native toolkit while maintaining a single, maintainable executable. It’s a pragmatic patch for a decade-old problem: the cognitive load of switching between ls and dir , or grep and findstr . While some might see this as just anot...

A newly disclosed vulnerability dubbed CIFSwitch is forcing Linux administrators to take a hard look at their file sharing configurations. The flaw, which reportedly grants root access across multiple distributions, exploits a previously unnoticed interaction within how certain Linux environments handle specific network file system requests. For platform engineering teams and self-hosters alike, it’s a stark reminder that foundational infrastructure components—the ones that sit quietly in the background for years—remain prime targets for critical escalation paths. While the technical specifics center on how the system processes CIFS/SMB mounts under specific conditions, the operational reality is more pressing. Many organizations treat internal file shares as low-risk zones, often deploying them with default settings behind a firewall. CIFSwitch demonstrates that when a service requires elevated privileges to mount and manage networked file systems, any edge case in its input validati...