All Posts

This is written for two people who rarely read the same document: the CEO who wants to know whether the company is exposed, and the CISO who has to answer for it. It is Monday, the week's breach headlines are already sorted, and there is an uncomfortable pattern in them that neither a bigger budget nor a bigger brand fixed. We will give you the pattern, the receipts, the two things that genuinely changed this week, and the short list of questions a board should ask before Fri

Here is the honest disclosure up front, because receipts cut both ways: we do not have a receipt on this one. We did not flag it early, we have no prior post to point at, and nobody has confirmed in-the-wild exploitation yet. This is a patch-ahead warning, not a victory lap. We are writing it anyway, because the thing that makes UniFi dangerous is the exact thing that keeps it off everyone's threat model — people trust it, and the trust is the vulnerability. On July 2, Ubiqui

This is a loop-closing post, so let us close it honestly, with both halves of the ledger showing. On July 2 we published a patch-ahead warning on Adobe's emergency ColdFusion bulletin and called exploitation "a when, not an if." Later that same day the technical details of CVE-2026-48282 — an unauthenticated path-traversal bug that hands a stranger remote code execution on ColdFusion 2025.9, 2023.20, and everything older — went public, and exploitation began within two hours,

On May 16 we published a breakdown of five Cisco Catalyst SD-WAN Manager vulnerabilities that CISA had just piled onto its Known Exploited Vulnerabilities catalog, several of them on the same day. Our headline said you could chain them to go from an anonymous HTTP request to owning every router in the fabric. Our closing math was blunt: "The total wire-level footprint of a complete fleet compromise is a handful of HTTP requests. Two minutes from anonymous stranger to controll

Earlier today we wrote about Progress telling ShareFile customers to physically power down their on-premises Storage Zone Controllers over a threat it would not name. The obvious next question is the one that decides how much this matters: who actually runs these things? Because a vulnerability's blast radius is not the CVSS score — it is the profile of the people holding the exposed box. And ShareFile's profile is a specific, soft one. We also went and checked our own brand-

When a software vendor tells you to apply a patch, that is a Tuesday. When a vendor tells you to physically walk over to the server and pull the power — and cannot tell you which vulnerability, cannot give you a CVE, and does not have a fix — that is something else. That is the vendor knowing something they are not saying. Progress Software just did exactly that to its ShareFile customers, and the silence is the story. What Progress said Progress emailed customers running on-

Yesterday we told you about the jscrambler npm compromise and made a point of it: we caught the indicators, hand-ingested them, and the free check-package tool we shipped could then block the stealer for any AI agent that asked. That was us doing the work by hand. Today the same class of attack hit a different target, and this time we did not have to lift a finger. The deny-list already had it. That is the version of this that actually scales, so it is worth showing you. What

Let us put our cards on the table before we say a word about this one. We build on Claude every day. Claude is the model we partner with, the one we trust to hold against prompt injections when other agents fold, the one whose reasoning runs underneath most of what we ship. So when we say what we are about to say, understand that it comes from someone who wants Anthropic to win, not someone looking for a reason to swing. This is a family critique, not an ambush. And the criti

We are going to be straight about the timing on this one before we say anything else: Oracle E-Business Suite has been under active attack since late June, and we are writing about it in mid-July. This is a catch-up, not a scoop. But the reason it is worth catching up on is the pattern it completes, and the pattern is the thing that should change how you treat your ERP. The flaw CVE-2026-46817 is a critical vulnerability — CVSS 9.8 — in the Oracle Payments module of E-Busines

On July 7, CISA added CVE-2026-55255 to the Known Exploited Vulnerabilities catalog and ordered federal agencies to patch it by the end of the week under Binding Operational Directive 26-04. The vulnerable product is Langflow — the open-source, drag-and-drop builder people use to wire up AI agents and LLM workflows. This is the fourth time Langflow has landed on the KEV list. We want to talk about why that number matters more than any single one of the flaws. What the new fla

Yesterday afternoon we shipped a free tool called check-package: a single call your AI coding agent can make to ask us whether a dependency is malicious before it runs the install. One URL, no key, backed by an OSV-curated deny-list of a quarter-million named-bad packages. We said the whole point was to collapse the friction between "a supply-chain attack exists" and "your agent is protected from it." Then, roughly a day later, the world handed us a test. What dropped On July

Earlier today we published a field map of 2026's novel attack chains, and the argument underneath it was that the AI agent has become the pivot point of the modern attack: a privileged process, sitting on your secrets, that acts on instructions it reads from its environment. One of the cleanest ways to weaponize that is the supply chain — get the agent to install a poisoned dependency, and you are inside. HalluSquatting does it by exploiting the fact that models invent packag

If you read enough incident writeups in a row, you start to see the shape underneath them, and the shape changed this year. For most of computing history, an attack chain meant breaking in: find a flaw in a server, exploit it, get a shell, move laterally. The novel chains of 2026 — the ones that are genuinely new, not last decade's technique in a fresh wrapper — mostly do not break in at all. They convince a program that already has your keys to hand them over. That program i

Late last month an anonymous researcher operating as "bikini" tried to shock the security world, and largely succeeded. They published the Exploitarium — a single repository of more than 130 unpatched proof-of-concept exploits spanning 22 software projects, dumped all at once with no notification to a single vendor. It rocketed to number five on Hacker News. The premise of a mass drop like this is fear: when a researcher publishes a working exploit without telling the vendor,

Today the Australian Cyber Security Centre — a national government cyber authority — warned the world about a large-scale global campaign exploiting content management systems: WordPress and a stack of its plugins, Craft CMS, MaxSite, MetInfo, Joomla JCE. The method is the oldest one in the book for this target class: exploit a file-upload or code-execution flaw, drop a webshell, and settle in for persistent access, credential theft, and lateral movement. It is a real, seriou

Almost every crypto theft we write about is an operational failure: someone got phished, cloned a webshell, installed a malicious wallet extension, approved a transaction they shouldn't have. Ill Bloom is not that. Ill Bloom is a story about mathematics, and it is the scariest kind, because the victims did nothing wrong. They generated a wallet, wrote down the twelve or twenty-four words, and stored them properly. The problem is that the words were never random in the first p

Three things off this morning's sweep that did not merit their own headline but should not go into the weekend unremarked. One of them is on a beat we have been standing watch over all year. GigaWiper: an Iranian destruction kit disguised as a Windows update This is the one to actually care about, because it is our beat and because it is built to end machines, not ransom them. Microsoft Threat Intelligence documented GigaWiper, a Go-based Windows backdoor that is really three

Here is the newest way to rob a codebase, and its cleverness is that it hides in the one place your automated reviewer refuses to look. Researchers call it Ghostcommit, and it is the third time in eleven days we are writing the same underlying story: the AI agent gets fooled by a blind spot in how it reads the world, and the tell is whether the model understands what it is actually looking at. This time the blind spot is an image file, and — for once, and worth saying plainly

It is Friday night, Central time, and three separate crews are working three separate doors into your estate. That is not a coincidence of the calendar — extortion operations prefer the weekend precisely because your team is not watching. Here is what is live right now, why each one matters, and what to do before Monday. Held to about ninety-five percent, and where the work is someone else's, it is credited. 1. ShinyHunters are still inside PeopleSoft — and there's a second C

There is a category of attack that only became possible when developers started letting AI assistants fetch things for them, and this week researchers gave its nastiest form a name: HalluSquatting. The premise is almost too simple. Large language models, asked to write code or set up a project, confidently invent the names of packages, repositories, and plugins that do not exist. They do it a lot — the new research measured hallucination rates as high as 85 percent when cloni