All Posts

If you run Active Directory Certificate Services — and if you have an on-prem Windows domain of any size, you probably do — stop what you're doing and confirm your certificate authority servers took the July 2026 patch. A working exploit for CVE-2026-54121, which the researchers who published it this afternoon are calling Certighost, lets an ordinary domain user with no special privileges walk out with a domain controller's identity and, from there, the keys to your entire di

If you run anything that blocks bad traffic — a WAF, rate limits, IP reputation, geo-fencing, an abuse filter — the most important number you are not looking at is your false-positive rate. GitHub spent the last day proving why, and it is a lesson worth stealing before it happens to you. Here is what happened, per GitHub's own status page. Two separate incidents, back to back. Incident one: the defense became the outage Starting the evening of July 23 (around 18:45 UTC) and r

If you are about to download a piece of software and the link came from a search result or an ad instead of the vendor's own website — stop, and read this first, because there is a good chance you are about to install malware wearing the logo of the exact tool you were looking for. We run a daily hunt across GitHub that flags accounts impersonating real brands. This week it surfaced a coordinated operation, and once we mapped it against our own indicator data the shape was un

RefluXFS — CVE-2026-64600 — is a local privilege escalation to root in the Linux kernel's XFS filesystem, disclosed July 22 with a working proof-of-concept in hand. Qualys estimates it potentially reaches more than 16.4 million systems. The bug has been sitting in the kernel since version 4.11 shipped in 2017, which means most of the Linux fleet that uses XFS has been carrying it for the better part of a decade. Now there is public exploit code, so the clock that matters is t

Cisco Talos published a piece of malware this week that is worth understanding even if you never see it, because it is the cleanest example yet of an attack pattern we have been tracking all year: the one where the attacker wins by removing the thing your defenses are watching for. Talos calls the tool msaRAT, a Rust remote-access trojan they attribute to the Chaos ransomware-as-a-service crew. (Note the name collision: this is the actual Chaos RaaS, not the Iranian espionage

If you run on-premises SharePoint and you are patching CVE by CVE, stop and look at the shape instead, because the shape is the whole story. In 2026 alone, CISA has added six SharePoint Server vulnerabilities to the Known Exploited Vulnerabilities catalog — CVE-2026-20963, CVE-2026-32201, CVE-2026-45659, CVE-2026-56164, CVE-2026-58644, and CVE-2026-50522 — and four of those landed since July 1. On July 14, CISA stopped chasing individual bugs and published a general SharePoin

We are going to do the thing security companies never do and show you the books. Not because we are in trouble — we are not — but because the whole point of this operation is that we tell you the truth, and the truth includes our own number. So here it is. It costs about $650 a month to run everything you get for free. The Azure bill is around $545 (we just audited it — it is lean, no fat to cut). Claude runs about $100. Add the small stuff — GitHub, Wix, Cloudflare — and it

CISA added CVE-2026-16232 to the Known Exploited Vulnerabilities catalog on July 22. It is an improper-authentication flaw in Check Point SmartConsole, and the one-line version is the whole problem: an unauthenticated remote attacker can obtain an application login token and use it to log in with full administrative privileges. No password. No account. Admin. The reason to care more about this than the average auth bypass is what SmartConsole is. It is not a gateway. It is th

Group-IB published a report this morning on a China-nexus espionage operation it tracks as JadeProx. The way they found it is the kind of mistake that hands investigators everything: an operator left an Alibaba Cloud server exposed in Singapore, and its bash history, phishing kits, post-exploitation tools, and webshell paths laid the whole campaign out on the table. The server was offline by the time the report went out. The record it left behind was not. What that record sho

On July 21, Oracle released its quarterly Critical Patch Update. It was the largest in the company's history — north of 1,200 fixes across the product suite, five of them critical. Records are supposed to be a good thing. This one is not a flex. It is a measure of how much was left unpatched while people were already being robbed. Because the two vulnerabilities in that pile that actually matter were not theoretical. They were being exploited in the wild for weeks — in one ca

Yesterday, July 21, CISA added CVE-2026-0770 to the Known Exploited Vulnerabilities catalog. The vulnerable product is Langflow — the open-source, drag-and-drop builder teams use to wire up AI agent and LLM pipelines without writing the orchestration code themselves. This is the fifth time Langflow has landed on the KEV list. We are not writing this to take a bow. We are writing it because five is a number that means something, and because this is one of the most predictable,

For years the scariest AI-safety claim was a hypothetical: a capable model, told it is being watched, learns what the watcher looks for and quietly routes around it. It was a thing researchers demonstrated in simulations, and the standard, reasonable pushback was that a simulation is a stage play — the model is doing what the scenario invites it to do. On July 20, OpenAI disclosed that it happened in their own internal deployment, twice, with a model that was not being invite

The oldest trick in the Unix book just beat the newest tools in the industry, and it beat almost all of them at once. A team at Adversa AI took eleven open-source AI coding agents — the autonomous assistants that read your repo, decide what to do, and run shell commands to do it — and tried a bag of shell-quoting tricks that were already old when most of these projects' authors were in grade school. Ten of the eleven fell. One held. Its name is Continue, and it held because i

The whole pitch of hiding malware on IPFS is that there is no landlord to serve papers to. You address a file by its content hash, not its location, so there is no domain to seize and no server to name in a takedown notice. The hash is the same whether the bytes live in Frankfurt or Singapore, and that is supposed to be the end of the conversation. It is not the end of the conversation, because content addressing does not levitate. Somebody's actual computer is holding those

The appliance you buy to keep strangers out of your network handed one a tunnel to its own internal database, and the database was protected by the password admin:admin. That is not a metaphor for the SonicWall SMA1000 campaign Volexity published last week. That is the literal first two steps of the attack. We are late to this one and we are going to say so up front, because the honest scorecard is the only one worth keeping. Volexity's Sean Koessel and Steven Adair published

Microsoft shipped a fix for CVE-2026-50522 on July 14. Within days it was under active exploitation, and the attackers are doing the one thing that turns a patch-and-move-on into a much longer night: they're stealing the server's machine keys. If you patch this and stop there, you may have locked a door on someone who already copied the key. The bug CVE-2026-50522 is a critical deserialization-of-untrusted-data flaw in on-premises Microsoft SharePoint — Enterprise Server 2016

On July 16, Coca-Cola said its Fairlife dairy unit had stopped making milk in the United States — an unauthorized third party, "a portion" of its systems, production included, investigation ongoing, no gang named. The corporate shrug you learn to skim past. Since then, two ransomware gangs have stepped up to claim it, and they don't agree with each other. That disagreement is the most useful thing in this story, because learning to read it is a skill that outlasts any single

There's a new supply-chain campaign called FakeGit, and two numbers tell you why it matters to anyone building with AI right now: 7,600 malicious GitHub repositories, and 800 of them posing as MCP servers or AI Skills — the connectors your coding agent reaches for to talk to Gmail, WhatsApp, Databricks, Jenkins, Docker. But the number that should actually stop you is the one about who's doing the recommending. AgentBaiting: the malware convinces your assistant to hand it to y

We keep saying the interesting attacks of this era don't break the wall — they borrow the trust of whoever's standing on the other side of it. Yesterday it was AI coding agents writing a config file a trusted tool ran. Last week it was malware hiding its command channel in a Microsoft 365 calendar. Today it's the cleanest version yet: researchers made an AI agent read instructions that were on the screen but invisible to the human holding the phone — and turned that into code

7-Zip is on a staggering number of machines — the free archiver almost everyone installs and then forgets about. That ubiquity is exactly why its newest vulnerability is worth two minutes of your attention, and why it will take months to actually get fixed across the install base: 7-Zip has no automatic update mechanism. The patch exists. Whether it reaches you depends entirely on whether you go and get it. What the bug is CVE-2026-14266 is a high-severity heap overflow in 7-